* llama: share the nextn tensor flags between models
Follow-up of the TODO in glm5-next: move the trunk-only and MTP-only
detection that each model copied into a nextn_flags helper of
llama_model_base. It probes the first trunk layer and the first NextN
layer, and adds TENSOR_SKIP when MTP is not loaded. qwen4exp probes
hc_attn_norm since it has no attn_norm.
deepseek4, nemotron-h, qwen35, qwen35moe, qwen3next and qwen4exp now
also accept a trunk-only file, like the other models.
* llama: avoid capturing structured bindings in the nextn flags
Lambdas that capture structured bindings need C++20, and GCC 15
rejects them under -Werror, so the models read the trunk and MTP
flags into plain variables.
The gather path attended over the selected latents with a plain
matmul and softmax. It only ran with n_ubatch <= 16, and the flash
attention backends now skip the masked rows through n_kv_max, so
the scatter path covers every case.
Drop the gather flag, the gathered attention branch and
gather_mla_rows. set_input_kpool always maps padding to the n_kv
sentinel, and the slot mask becomes sel_mask since only the
scatter reads it.
* ggml: fix CLAMP on non-contiguous views (CPU, CUDA)
CUDA clamped ggml_nelements values flat and ignored the view strides.
CPU addressed row j as j*nb01 and ignored nb02/nb03. Both now follow the
strides of dims 1..3; CUDA supports_op requires contiguous rows, like
Metal. test_clamp gains a non-contiguous view case.
* cuda: clamp kernel uses fastdiv for the view strides
* llama: re-pool each shared k-pool rep once
With shared cells every pool is re-pooled, and since the pooled keys
are always scattered, the pools a seq_cp shares between sequences
wrote the same rep row from several scatter entries, a data race on
the CPU backend. Mark each rep once: the sharing sequences read the
same row through pool_cells.
* llama: assert whole-sequence seq_cp in the hybrid idx memory
The recurrent state is always copied whole whatever the range, and a
k-pool cell shared by a partial copy could carry two pool groupings
with a single pooled row. Every caller copies whole sequences, so
reject partial ranges instead of supporting them.
* llama: drop the k-pool cache_safe mode
With whole-sequence seq_cp, sequences sharing cells share their pools
too, so the pooled row of a shared rep is valid for all of them. Mark
each rep once in every ubatch instead of re-pooling everything while
cells are shared, which removes the sharing scan and the stale-all
workarounds in seq_rm, state_read and state_drop. seq_cp now only
stales the destination.
* cuda: stage the lightning indexer queries in head passes for MUSA
MUSA archs 21 and 22 cap static shared memory at 28 KB, and the tile
kernel staged the queries of all four heads next to the key tile for
33 KB. The queries are now staged in passes of
LIGHTNING_INDEXER_TILE_HEADS_PER_PASS heads: two on MUSA for 25 KB,
four elsewhere where the single pass folds to the previous kernel.
* cuda: use the vector lightning indexer kernel on MUSA
Address review from am17an: the tile kernel stays off MUSA, whose archs
21 and 22 cap static shared memory at 28 KB, below the 33 KB the tile
needs, so MUSA keeps the vector kernel it ran before. This replaces the
head passes, CUDA and ROCm run the merged kernel unchanged.
* cuda: tile the lightning indexer over keys and tokens for 4 heads
With too few heads for a wmma tile, a block scores 64 keys against 8
tokens: the keys are staged once in half precision, the queries one
head at a time, and each thread owns one key for two tokens, so no dot
product needs a cross thread reduction. Batches smaller than a token
tile keep the vector kernel. test-backend-ops measures 4 heads.
* cuda: multiply the lightning indexer tile in float
Address review from am17an: the half2 products overflow once a single
q * k exceeds the f16 range. The queries stay in float in shared memory
and each half2 of keys is widened once for both tokens, so every
product and sum is computed in float.
* cuda: widen each lightning indexer key once for all heads
The tile kernel stages the queries and weights of every head at once,
so each key element is widened from half once and feeds all heads,
with a single barrier. F16 keys are copied into the tile without a
float round trip. Keeping the keys in float in shared memory measures
slower, the occupancy drops.
* cuda: stop the lightning indexer tile from spilling registers on ROCm
Each thread of the tile kernel now scores two keys for a single token,
so a warp shares its token and the query reads are broadcasts: six
shared reads per element pair instead of nine for the same products.
The inner loop is unrolled by 8, which keeps gfx908 at 63 VGPRs with no
spill where the fully unrolled loop needed over a thousand, and makes
the kernel 36x faster on an R9700 and slightly faster on CUDA.
* log, server: make router child lines carry their own colors
The logger writes the color reset after the trailing newline, so the
reset opens the next line. On the shared pipe of a router child it lands
in front of the next state command, which the router then misses, and
the line break that works around it shows up as an empty log line on
every progress update.
The reset now goes before the trailing newlines, so every line is self
contained and the command goes back to its plain framing. The router
passes its effective color setting to its children, whose output ends
up in its terminal, and leaves that option out when comparing presets
on reload.
* log: enable ANSI colors on the Windows console
A Windows console renders ANSI sequences only in virtual terminal mode,
which nothing turns on for the logger, so llama-server prints raw escape
codes on the Windows 10 console while llama-cli, whose console code
enables it, shows colors. The logger now enables virtual terminal mode
on stdout and stderr when it turns colors on, and keeps colors off when
a console cannot render them. Pipes and files take the sequences as is.
* server: separate the router child commands from its logs
The child sent its state commands on the same pipe as its logs, so the
router had to pick them out of the log stream by a line prefix, and any
unterminated write in front of a command made the router miss it. This
resolves the TODO at the spawn that called for splitting stdout and
stderr.
The child now keeps stdout for the commands and points everything else
written to stdout at stderr, before anything is written. The router
reads both pipes, handles the commands from stdout and forwards stderr
as the log, and warns about any other line on the command pipe.
* server: address review from ngxson
The single server_child is now created first in the entry point and its
constructor keeps stdout for the commands, so the stream is a member of
the instance instead of a static, and init() is gone. The instance is
passed down to the server, while the CLI entry point creates its own.
* Update tools/server/server.cpp
---------
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
build_rs gathered the extra states (n_rs - n_seqs rows) with their own
get_rows. The worst-case reserve has n_rs == n_seqs, so that node was
sized at zero rows, and any ubatch whose cells are not contiguous forced
a graph reallocation at an unchanged node count, which aborts under
GGML_SCHED_NO_REALLOC.
A single get_rows now gathers the n_rs states: the ubatch states and the
extra states are views of it, and its size only depends on n_rs, which
the reserve already sets to the maximum. A custom getter (mamba ssm_scan)
gathers from the second state, so a single sequence ubatch copies no
state. The views are built once per graph in the input to keep the host
overhead of the graph unchanged.
* qwen4exp : halve the indexer score memory
The indexer scored all heads in one product and rectified a copy of it,
so two [n_pool, n_idx_h, n_tokens] f32 tensors were live at once, the
largest buffers of the graph at long context. Each head now gets its
own product, rectified and summed in place into one [n_pool, n_tokens]
score.
* qwen4exp: let the allocator reuse the indexer score buffers
Address review from CISC: use plain ggml_add and ggml_relu in the
indexer head loop. The graph allocator already runs them in place when
their source has no other consumer, so the _inplace variants are not
needed. The compute buffer and the speed are unchanged.
* cuda: support 4 heads in the lightning indexer
Dispatch 4 heads to the vector kernel, too few for a wmma tile, and
accept them in supports_op. test-backend-ops covers 4 heads.
* metal: take the lightning indexer head count as a function constant
The kernel reads the head count from a function constant and zero fills
the last head tile, so any head count runs and 64 heads is unchanged.
* qwen4exp: compute the indexer score with the lightning indexer
Address review from am17an: the unweighted sum of the rectified head
scores scaled by 1/sqrt(head_dim) is the lightning indexer with every
head weight set to that scale, so the indexer calls
ggml_lightning_indexer on the pooled keys with an f16 pool mask. The
keys are read once for all heads and no per head score is
materialized.
* vulkan: tile the lightning indexer over keys and tokens
A workgroup scores 64 keys against 8 tokens: the keys are staged once
in shared memory, the queries one head at a time, and each invocation
owns one key for two tokens, so no dot product needs a cross invocation
reduction. The subgroup variant and the flat dispatch are gone, the grid
is keys x tokens x streams.
* vectorize vulkan loads and use fp16 dot product
---------
Co-authored-by: Ruben Ortlam <rortlam@redhat.com>
The sparse indexer mask is built with a set_rows scatter. Padded pools,
absent sequences and missing tail cells all pointed to the same n_kv
sentinel row, and invisible pools picked by top_k to fill the selection
overlap the tail cells of the token, so several CPU threads wrote the
same element (ThreadSanitizer data race in the sanitize CI).
Allocate the slot mask for both selection paths and route every dead
slot to its own dump row n_kv + slot. Live slots address disjoint cells,
so the scatter indices of a token are unique.
* cli: exit on stdin EOF and drop the console wide Ctrl+C broadcast
On Windows the simple input reader sends CTRL_C_EVENT to every process
attached to the console when stdin reaches EOF, killing unrelated
processes such as a supervising agent. The CLI only stopped on EOF
because of that self inflicted SIGINT; on POSIX, and with the advanced
reader, it spins forever printing prompts.
Drop the broadcast so both platforms just return an empty read, and
treat an empty read as EOF in the chat loop and the model selection,
since a submitted line always ends with a newline.
* cli: keep the newline of a trailing "/" and stop mtmd-cli on EOF
A lone "/" came back as an empty read and was taken for EOF, and
mtmd-cli only stopped on EOF through the removed broadcast.
The fixture recycles its two blocks over 8 cache slots, so the fp16
error builds up past the 1e-4 NMSE bound on the Vulkan T4 and WebGPU
jobs of Models Backend. Two l-cycles keep every branch of the cycle
loop and halve the error.
* cpu: accept BF16 in src1 of mul_mat
ggml_conv_1d_dw builds its im2col in F32 when the kernel is BF16, then
calls ggml_mul_mat(im2col, kernel), which puts F32 in src0 and BF16 in
src1. The CPU backend refused that combination, so it was reported as
unsupported on every backend and never compared against anything.
Widen BF16 into the F32 work buffer, next to the existing packing of F32
into vec_dot_type. This is the arithmetic the Metal mat vec kernel
already uses, both operands promoted to float and accumulated in float,
so the two agree exactly rather than approximately.
Cover it with a conv_1d_dw test over F32, F16 and BF16 kernels, plus
three mul_mat cases with BF16 in src1.
* vulkan: reject BF16 in src1 of mul_mat unless src0 is BF16
supports_op only checked the src1 type for non contiguous tensors, so
a contiguous BF16 src1 was accepted and the pipeline lookup asserted.
The only BF16 src1 path is the BF16 x BF16 multiply, every other src0
type now reports the op as unsupported and the scheduler keeps it on
the CPU.
The BF16 kernel case of the conv_1d_dw test needs the f32 x bf16
mat vec variants of the Metal backend, which land separately.
* openvino: serve GET_ROWS on a weight view from the base Constant
Resolve view_src when collecting weight Constants so a view over a
quantized weight no longer becomes a dynamic typed Parameter, and fold
the row offset of the view into the gather indices instead of slicing
the dequantization subgraph.
* openvino: lift the quantized GET_ROWS view rejection
The supports_op rejection of a quantized src0 view with a nonzero
offset keeps the vs0 GET_ROWS cases of #28253 away from OpenVINO.
The weight view now resolves to the base Constant with the row offset
folded into the gather indices, so the rejection goes away.
#27773 adds the glm5-next arch without its rows in the Metal fusion
baseline, so test-fusion --check fails on it. The rows come from
test-fusion --record on an M5 Max, and --check passes 270/270.
Without CUB (HIP, MUSA) argsort ran the bitonic kernel with one thread
per padded column, so any row above 1024 entries launched an invalid
block configuration. Each thread now owns several columns, every stage
of the network runs all owned columns before the barrier, and the block
is capped at 1024 threads. Shared memory becomes the only bound, which
supports_op checks against the device instead of a fixed 1024.
Rows up to 1024 run the same work as before. Bit-exact with the CUB
path on rows of 2048.
* models: pad on the left with ggml_pad_ext
The Parakeet, LFM2-Audio, Granite Speech and Gemma 4 audio encoders
build a left padding as a right pad followed by a roll, and DFlash2
concatenates a zero filled block in front of the previous tokens.
ggml_pad_ext does both in one node now that every backend supports a
left padding. The Gemma 4 audio embeddings are bit identical.
* models: skip the DFlash2 taps that only read padding
A tap at or past block_size shifts every row out of the block, so its
term is zero. The loop runs min(kernel_size, block_size) taps.
* metal: support left and circular padding in GGML_OP_PAD
Align Metal with CPU, CUDA and Vulkan: shift the source coordinates by
the left paddings, wrap them around with the same wrap_around when
circular, and read the source through nb00, which also fixes a right
padding of a permuted source. A test case covers it.
Drop the f32_4 kernel: its selection is disabled as slower, and it
fails two pad cases once enabled.
* metal: use a function constant for the circular pad variant
Address review from ggerganov: replace the bool template with FC_PAD,
as FC_upscale_aa does, so the pad kernel is compiled once and
specialized per pipeline.
* vulkan: read the batch stride of an in place src0 from nb[2]
A dim01 contiguous tensor can still be a view whose batches are
strided by more than ne[1] rows, the first rows of a KV cache for
example. Both the mat-vec and the matrix paths read such a tensor in
place but passed ne00*ne01 as the batch stride, so every head past
the first read the wrong rows. The same applies to src1. The stride
now comes from nb[2] whenever the tensor is used in place; the value
is unchanged for a contiguous tensor.
test-backend-ops gets an m_v parameter on test_mul_mat, the number of
rows of a in memory, and two cases at the shapes of a decoder self
attention over a cache.
* vulkan: size the in place A and B ranges by their strided extent
The matrix path bound src0 and src1 to the shader with a range of
elements times type size, which ends before the batches of a strided
view. Pipelines with bounded access read zero past that range, so the
same view that the mat-vec path already handles gave wrong results
on Intel and on NVIDIA without coopmat2. The range now comes from
ggml_nbytes when the tensor is read in place.
* vulkan: address review from jeffbolznv
Bind the in place A and B of the matrix path with ggml_vk_subbuffer,
which spans to the end of the buffer, so a strided view is in range
without computing its extent.
mul_mat_id reads the batch stride of an in place src0 and src1 with
the same helper as mul_mat. test_mul_mat_id gets an m_v parameter,
the number of rows of as in memory, and a case whose experts are
strided by more rows than it uses.
* vulkan: read the batch stride of an in place src0 in mul_mat_vec_id
The single token path of mul_mat_id passed ne00*ne01 as the batch
stride of A, so a strided expert view read the wrong rows. The stride
now comes from ggml_vk_batch_stride like the other three paths, and
src1 follows the same rule.
test_mul_mat_id gets a single token case over the strided view.
* vulkan: address review from jeffbolznv
The batch stride of an in place tensor is taken from nb[2] as
nb[2] / type_size * block_size, which holds when nb[2] is padded and
not a multiple of nb[1]. A test_mul_mat case with a padded batch stride
covers it.
* vulkan: keep the A and B ranges exact in mul_mm
The quantized A loads of mul_mm carry no row bound and rely on the
descriptor range to read zeros past the last row of a partial tile.
Binding A and B up to the end of the buffer let those tiles read the
leftovers of a previous node and hung the NVFP4 mul_mm on NVIDIA
without coopmat2. The range is the strided extent of a tensor read in
place and the staged size otherwise.
ggml_conv_1d_dw builds its im2col as f32 when the kernel is bf16, then
multiplies the two, so a depthwise convolution over bf16 weights asks
for kernel_mul_mv_f32_bf16, which was never instantiated. The base, the
_4 and the _short families are filled in next to their bf16 neighbours,
inside the same runtime guard, so a device without bf16 support is
unaffected.
* server: route every model load through the queue
A model loaded by the fast path has no queue entry, so tick() evicts
it at its LOADED transition before its own request is proxied. Every
load now joins the queue, whose entry protects the model until its
waiters leave.
* server: do not admit requests into a stopping model
A request for a model that is being stopped still sees it LOADED and
is proxied into the dying child. Such a request now joins the queue
and is served by the next instance. The stopping mark is cleared
under the same lock that sets UNLOADED, so no request can see a
model that is neither stopping nor unloaded while its child is gone.
* llama: read the SWA pattern as a period or a per-layer array
Add llama_model_base::load_swa_pattern(), which reads
sliding_window_pattern either as one flag per layer or as a period
expanded by set_swa_pattern(), and use it in every loader that reads
the key as a period.
These loaders silently ignored an array and applied their default
period, although the converters of olmo2, gemma3n and exaone4 write
arrays. The published GGUFs match the defaults, so their outputs do
not change. The loaders that already accepted both forms lose their
duplicated scalar-then-array block, and use their declared default
period when the key is absent.
* model-saver: write the SWA pattern and the MLA SWA geometry
Write sliding_window_pattern as one flag per layer, nextn layers
included, for every model using SWA. The array is never collapsed to
a scalar, since the loaders read a scalar as a period.
Also write the MLA key/value lengths and KV LoRA rank of the SWA
layers, required by dots3note.
This enables the saver for plamo3, gemma3, cohere2, cohere2moe,
olmo2, exaone-moe, afmoe, mimo2, spark2_5, muse-glimmer, mellum,
laguna, granite_swa, dots3note and maple, all passing the bit-exact
roundtrip of test-llama-archs.
Clang stores the modification time of the precompiled header sources
inside the header and refuses the header when they differ. A cached
header restored from another checkout carries the timestamps of that
checkout, so the build fails. The option covers the compilers ccache
treats as MSVC while they are clang underneath, clang-cl and the Intel
LLVM drivers.
The child writes its state commands on stdout while the logger writes
on stderr, and both share a single pipe. The logger emits the trailing
color reset after the newline of a debug, warn or error entry, so that
escape sequence has no newline of its own and the router reads it glued
in front of the next command. The line prefix check then fails and the
command is forwarded as a log line instead of being handled, which
leaves a finished download stuck in the downloading state.
Writing the command with a leading newline closes the pending line so
it always starts at a line boundary.
Walk the binding offset back until the distance to the tensor is a
whole number of blocks, so block quantized views get a valid element
offset in the shader.
The expected success table holds when the four requests enter the shared
pool together. On a loaded runner they are admitted tens of milliseconds
apart, the slot lifetimes overlap differently and the pool overflows
while a short request is still resident. The decode failure aborts every
slot, so a request the table marks as successful comes back with the
context error instead of its generation.
Such a request now passes on that error too, while any other status, a
different error or a truncated generation still fails the test.
Templates that default an optional variable to none and then test its
membership in a map hit an error, while the same expression is a normal
lookup returning false in Jinja. The undefined counterpart of this case
was already handled just above.
* tests: bind the L2_NORM batch count to a local
GCC cannot prove the loop fills norms up to the index read after it
while the bound is a class member, so it reports a maybe uninitialized
use. Reading the count once into a local restores the tracking.
* tests: initialize the L2_NORM batch array
The read after the fill loop is only provably defined once the array
carries an initializer, which GCC 12 requires on the aarch64 Release
build where warnings are fatal.
* mtmd: load the qwen3-tts code predictor proj_in as optional
The talker and the code predictor share the hidden size on the 0.6B
checkpoints, so the reference builds no small_to_mtp_projection and
the conversion emits no tensor for it. The graph already falls back
to identity when the weight is missing, the loader now agrees.
* mtmd: keep the qwen3-tts code predictor ffn_down in F32
The code predictor carries a massive activation: its layer 2 FFN
intermediate peaks around 1.5e5, well past the 65504 ceiling of F16.
mul_mat casts its input to the weight type, so an F16 ffn_down turns
that peak into inf, the residual follows, and the next rms_norm yields
NaN. Reference forward in float32 gives 145109 against 145396 measured
in the graph.
get_prev_tokens() rebuilt a (seq, pos) -> token hash map on every
ubatch by walking all used cells, while llama_kv_cells already keeps
an ordered index of the positions of each sequence in seq_pos, updated
on every cell mutation to serve seq_pos_min() and seq_pos_max().
The index now stores (pos, cell) pairs in a std::set instead of a
position -> count map, so a repeated position (cache reuse via rm + add,
vision inputs with shared positions) yields distinct entries and the
removal of a cell erases its own pair. The new seq_pos_tok_le() returns
the token of the cell at the largest position <= p in logarithmic time,
which is exactly what the old window lookup and its M-RoPE gap fallback
computed together.
get_prev_tokens() shrinks to a direct lookup per (token, offset) and
for_each_token_in() goes away with its only caller. The kv-cache keeps
no n-gram logic of its own.
Measured on Qwen3.8-Flash-Next UD-Q4_K_XL at 71k context, alternating
two binaries with the first run discarded: tg 69.3 -> 72.7 t/s (+4.9%),
pp unchanged at ~2720 t/s, greedy output identical, needle retrieved.
MTP speculative decoding needs the target state to move back by the
number of rejected draft tokens. Without rollback support the context
is classified as SEQ_RM_TYPE_FULL and the server serializes the whole
recurrent state to host memory on every round, which costs more than
the drafting saves.
The recurrent cache already holds n_rs_seq + 1 snapshot planes and the
delta net writes its SSM state into them, but build_conv_state_at wrote
a single plane, so a rollback restored a convolution history that was
never captured. It now writes one snapshot per slot, each ending one
token earlier, for the delta net QKV convolution and for the PLE
convolution alike.
Measured on Qwen3.8-Flash-Next UD-Q4_K_XL with the standalone MTP
draft, n-max 3 and a single slot: decoding reaches 183 tok/s on code
and 144 tok/s on prose. The same branch before this change, where the
server falls back to checkpointing the state to host memory, reaches
123 and 83 tok/s, for 108 tok/s without a draft.
* qwen4exp: sum the indexer heads by slices
The head reduction went through a transpose and a sum_rows over ne[1],
which left sum_rows with ne0 = 4, one block per row for a four element
reduction, and the transpose copied the whole block by token surface
twice on the way in.
The heads are adjacent on ne[1], so each one is a strided view and the
sum is a short chain of adds.
RTX PRO 6000, Qwen3.8-Flash-Next UD-Q4_K_XL, fa on, 55k context, warm
runs on top of #28011:
prompt processing 2170 -> 2366 t/s
Generation is unaffected. The removed work scales with n_blocks by
n_tokens, so the gain grows with context and with ubatch size.
* qwen4exp: drop the redundant cont on the indexer query
rope returns a freshly allocated, contiguous tensor, so the reshape that
feeds the matmul does not need a copy. ggml_reshape_3d asserts
contiguity, so a layout that would need the cont cannot slip through
silently.
Greedy output is unchanged token for token.
Address review from @ggerganov
* ui: copy the displayed text of grouped agentic responses
Agentic sessions render as a single entry anchored on the first
assistant turn, whose content is typically just the first tool call,
so the copy button wrote an empty string to the clipboard. Derive the
text sections of the whole session and copy them joined, matching the
visible response. Plain messages keep the previous behavior.
* const
for_each_token_in tested all LLAMA_MAX_SEQ sequences for every used cell,
while a cell almost always belongs to one. The scan now stops once the
cell's own sequences have been seen. Same visit order, same callback
arguments, so behaviour is unchanged.
get_prev_tokens is the only caller, so this affects the n-gram path.
RTX PRO 6000, Qwen3.8-Flash-Next UD-Q4_K_XL, fa on, warm runs:
55k context generation 56.3 -> 74.3 t/s
132k context generation 33.6 -> 50.9 t/s
Prompt processing is unchanged, the scan is amortised over the ubatch
there. The gain follows the number of used cells, so it grows with
context and is invisible on short prompts.
The optimized path grouped warp lanes by token and required
warp_size % n_expert_used == 0, with a single hardcoded exception
padding 6 up to 8. Every other count fell back to the generic path,
which walks the tokens one at a time with a warp reduction per token,
for each of the n_expert blocks.
The lane group only has to divide the warp, and the loop body already
guards the padded lanes with iex < n_expert_used, so the padding
generalizes to the next power of two. The 6 -> 8 case and every count
already dispatched keep the exact same padding as before.
n_expert_used = 10 now reaches the fast path. Measured on
Qwen3.8-Flash-Next (512 experts, 10 used) at 55k context on an
RTX PRO 6000, warm runs with the first one discarded:
prompt processing 2334 -> 2600 t/s
Token generation is unaffected, since a single token leaves nothing to
walk. Other expert counts reach the fast path by adding their case to
the dispatch.
* ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE
__ARM_NEON only signals NEON availability. The __fp16 type also needs
the IEEE half format, implied on AArch64 but selected with
-mfp16-format=ieee on 32 bit Arm, where the compiler otherwise rejects
the type.
The guard keeps every toolchain that provides the type on the same code
and sends that one configuration to the generic lookup path.
* ggml-cpu: gate the NEON+FMA block on __ARM_FP16_FORMAT_IEEE
Both halves of the F16 section dereference __fp16, so armv7 with
neon-vfpv4 hits the same unknown type error. Without the IEEE
format the configuration now falls back to the scalar path.
Address review from @JonathanC-ARM
The route loads run ahead of the root layout script, so validateApiKey
read the settings store while it still held factory defaults and probed
/props without the stored key. initStores() now hands the same startup
promise to every caller and the chat loads await it before probing.
The one-time admin baseline no longer overwrites a key the user has
already set: on a first visit the config carries factory values only, so
a diverging key comes from the user and wins.
* doc: document MCP stdio servers and CORS defaults in the server README
The MCP arguments were listed but nothing explained what an MCP server
is or how to declare one. Cover the stdio transport, the config keys,
the tool naming, and add a POSIX shell echo server as a minimal
example.
Also document the CORS behavior: the default reflected origin, the
switch to localhost once tools are enabled, and the recommended setting
per deployment.
* doc: drop the inline MCP shell example from the server README
The example parsed JSON-RPC by hand and sat in a page people copy paste
from, into servers spawned with the privileges of llama-server. Point to
the specification instead.
Link the pull request that introduced the feature, and keep a short
mcp.json snippet so the table of configuration keys has a declaration to
refer to.
* ui: split the markdown rendering setting per surface
User content and thinking get their own toggle again, so turning off
markdown for a message leaves reasoning blocks formatted. Both default
to markdown. A stored renderContentAsRawText unfolds onto the user key
and is dropped from the config.
File mentions render as badges in the raw text path too, through a
narrow pass over [name](file://path) that leaves everything else
untouched.
* ui: let the rich chat input scroll past its max height
The contenteditable renderer caps its height with max-height but had no
overflow rule, so a long buffer overflowed into the input area wrapper
and got clipped by its overflow-hidden, leaving no way to reach the
bottom of the message. The textarea renderer scrolls natively and was
never affected.
* ui: apply the new lint and format config
* ui: move the render keys unfolding into the migration service
Address review from @allozaur: the settings store no longer rewrites
persisted config on load, the raw text toggle now unfolds onto the
per-surface render keys in migration.service.ts, next to the other
config migrations. The mention scanner flag and the directory path
suffix become named constants.
* server: add an ssh transport to the tools runtime
--tools-runtime ssh:<target> runs the built-in tools on a remote host,
where target is whatever ssh already resolves, a user@host or a config
alias, so no credentials live in llama.cpp.
Only build_argv and upload differ from the docker transport: the remote
shell re-parses the command line, so the argv travels through
shell_quote_join, and files go over scp with the same quoting on the
remote path. Authentication is key-based and the host key must already
be trusted, since the tools run without a console and any prompt would
hang them.
The target is validated before use. The spec can reach us from the
x-tool-runtime header, and a leading dash would turn it into an ssh
option, which is enough to run a command back on the host.
Nothing is created and nothing is reclaimed, so an ssh spec goes
straight to the tool call instead of through the container runtime.
Note that this is remoting rather than isolation: the tools can do
whatever the target account can do, and the isolation is whatever runs
them on the far side.
* server: support podman in the tools runtime
docker and podman expose the same run, exec, cp and inspect verbs with the
same argument order, so a single implementation drives both and the engine
is carried by the spec prefix: podman:<image> and podman-container:<id> sit
next to the docker forms.
tools_io_docker becomes tools_io_container and the runtime spawner becomes
server_tools_container_runtime, both holding the client binary chosen at
parse time. A single parse_container_runtime() resolves every spec, so
adding another engine is one string in the table.
make_tools_io() now rejects the spawning forms. The spec also reaches it
from the x-tool-runtime header, which is client controlled, and only the
runtime that owns a container is allowed to create one: a tool call can
attach to a running container, nothing more.
* ./build/bin/llama-gen-docs
* server: simplify the tools runtime and drop the file copy step
A server_tools_runtime base with one virtual spec() replaces the
container runtime and the bare spec string that ssh needed next to it,
so server_tools is back to a single pointer and neither setup nor the
handler tests which of the two is set.
write_file used to spill its content into a temporary file on the host
and copy it in, because run_subprocess had no way to feed a child. It
now takes an optional stdin payload and creates the parent directory
and the file in a single round trip through a shell in the isolate.
That removes the upload virtual and both implementations: no more
container cp or scp, no second binary on the host, no sftp subsystem on
the target, no predictable temporary in a shared tmp, and none of the
content reaching an argv the remote shell re-parses. It also fixes
write_file over ssh, which never worked: scp speaks sftp and takes the
remote path literally, so quoting it kept the quotes in the file name.
Writing the payload before reading the output relies on the child
draining stdin as it goes, which holds for cat, its only user today.
* ./build/bin/llama-gen-docs
* server: harden the tools runtime against argv injection and a stdin stall
Validate the container id from x-tool-runtime and --tools-runtime the
same way the ssh target already is, so an id shaped like an option
(docker-container:--privileged) is rejected before it reaches the
engine's exec command line instead of running against a hardened
container. Feed the child's stdin after the watchdog is armed, so a
transport that stalls mid-write is terminated at the deadline rather
than blocking the request forever.
Cover both guards and fix the unknown-scheme test, which used ssh: as
its example and now names a real runtime.
* tests: exercise the tools runtime tests on podman as well as docker
Follow-up #26507. The container runtime drives docker and podman
through one implementation, so parametrize the availability helper,
the container fixture and the attach test on the engine, and cover
both engine prefixes in the container id injection test. Each engine
skips on its own when it is not installed.
The spawn cleanup test stays docker only: it recovers the spawned id
from the container hostname, which docker sets to the short id and
podman rootless does not guarantee. Podman keeps its coverage through
the attach path.
* server: release the container handle before respawning
Follow-up #26507. create() writes over the handle it is given, so a
respawn after the container died on its own leaked the pipes and the
process handle of the previous one.
* server: trim the tools runtime comments
* server: read tool output as raw bytes and harden the runtime on Windows
The stdout pipe is read with read() instead of fgets(), so a chunk
can hold any byte, including NUL, and still streams as soon as data
is available. Past the size cap the pipe keeps draining so the child
never blocks on a full pipe. Both pipe fds are forced to binary mode
on Windows, where the CRT defaults them to text mode and translates
line endings in both directions. Stdin is now always closed after
the feed: the child reads a deterministic EOF, and the Windows
docker and ssh clients stop outliving their command on a stdin pipe
that never closes.
The attach form of --tools-runtime has no lifecycle to own, so it
becomes a static target validated once at startup. This removes the
subprocess that ran on every tool call and
serialized calls behind a mutex; a stopped container now surfaces
the engine's own error at exec time.
The cidfile path is passed as UTF-8, matching the encoding the
subprocess layer expects for the CreateProcessW command line, so
the spawn form works from a non-ASCII Windows profile.
The SIGPIPE note in server.cpp now names the tools runtime children
as well as the MCP ones.
* clean up comments
* less pollute global scope
* nits
* tests: name the container image after both engines
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Upstream merged the Windows argument quoting fix, the NetBSD build
fix and the chdir fallback for glibc older than 2.29, so pin the
vendored copy to a commit that carries all three and remove the
patch files along with the apply step in the sync script.
The new pin also brings the exec error report on glibc older than
2.24 and the ENOSYS mapping to a dedicated error code. Both are
additive and no caller inspects those values.
docker info only proves the daemon answers, so the Windows CI passes
the check and then dies trying to run a linux image. The hosted
Windows runners cannot run one: GitHub states the VMs are not enabled
for nested virtualization and will not be, since they already sit one
level deep and the hypervisor does not support more levels
(https://github.com/orgs/community/discussions/25491). Probing the
image itself skips those tests there, and pulls it before the server
waits for the container id.
The picker mounts whenever a cwd-aware builtin tool is enabled, so
it can open while file_glob_search is not served or was disabled by
the user. Every typed query then fired a search that could only
fail with a raw error.
Gate the debounced search on the tool state, the same way the
mention picker does, and show a message in place of the results
list that explains why search is unavailable. Manual entry with
Enter still commits a directory. The Browse button and the search
scope footer are hidden as well: Browse resolves the picked folder
name through file_glob_search, and the client-side toggle would not
stop that call.
* server: report the isolate working directory from get_info
Without an explicit cwd, get_info fell back to the server process
working directory even when a tools runtime was configured. That named a
host path no tool would ever run in, since an isolate starts in a
directory of its own.
It now asks the isolate for its working directory in that case, and
keeps the process one only when the tools run on the host.
* remove redundant comment
---------
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
The working directory chip showed up as soon as the server exposed any
builtin tool, so a server started with just get_datetime, or a user who
turned every filesystem tool off in the settings, still got a control
that nothing would read.
Tools now declare whether they resolve their paths and run against the
working directory, next to the write permission they already publish in
the /tools listing. The WebUI shows the chip and enables the /cwd
command only when at least one such tool is both served and left
enabled.
get_output runs the waveform work the pipeline defers to it, from a
single trailing window to a full pass depending on the model. Measuring
it keeps the reported total and the audio to process ratio honest.
* feat: Add contenteditable tokenizer for badge/code-chip chat input
* feat: Add source-space undo/redo history for the rich input
* feat: Split text glued to a closing code fence onto its own line
* feat: Add ChatFormContenteditable rich input renderer
* feat : wire the contenteditable into ChatForm with auto-switch gating
* base : slash-command/misc foundation - model icon and focus-selector constants
* feat : slash-command picker and command parsing helpers
* refactor : wire command and @-mention pickers into the chat form
* ui : improve model selector keyboard navigation and load/dismiss
* feat: Unify markdown/raw-text rendering under one setting with migration
* fix: Misc fixes - tool-call subtitle, assistant wrap, progress guards
* feat: Clamp and style numeric settings inputs from registry bounds
* base : @-mention picker foundation - glob search, picker nav, highlight
* feat : @-mention file/folder picker and mention badges in message bubbles
* fix: Imports
* feat : wire the @-mention picker into the chat form
* fix: Bound the glob-search result cache key and prune stale entries
The reference implementation has two mutually exclusive prompt layouts.
In non streaming mode the prefill carries the whole utterance text plus
tts_eos summed with codec_pad, and the trailing text hidden collapses to
a single tts_pad row. In streaming mode the prefill carries only the
first text token and the trailing rows stream the rest of the text
followed by tts_eos.
The pipeline built the non streaming prefill but the streaming overlay,
so the talker saw the utterance a second time during generation and read
it twice before emitting codec_eos.
The overlay is now the single tts_pad row that matches the prefill.
* ui: read model modalities from the router model list
The router advertises input modalities for every model, loaded or not.
Reading them at list build time lets the UI accept image and audio
uploads for a model selected through ?model=, which has no /props yet.
* enum
* ggml: add ggml_build_forward_order
ggml_build_forward_expand marks the tensor and all its ancestors for
compute, so using it as a pure ordering hint (keeping q, k and v
together) defeats ggml_build_forward_select: the unselected branch is
forced to run with inputs that were never uploaded. In the mtmd audio
graph this makes GEN_WAV calls execute the GEN_CODE branch with a
stale inp_code0, hitting the get_rows bound assert on CPU.
Add ggml_build_forward_order, which inserts nodes without the compute
flag; the flag is restored when the branch is actually selected.
Switch the q/k/v hints in clip_graph::build_attn to it.
* nit: reduce comments (AGENTS.md)
* server: don't walk Windows junctions in file_glob_search
std::filesystem reports a junction as a plain directory, so the symlink
guard misses it and a junction pointing back at an ancestor is walked
until the path length gives out
read the reparse tag and treat a symlink and a mount point as links,
leaving any other reparse point walkable so cloud placeholders and dedup
stubs still get searched
look junk directory names up case insensitively on Windows, where NTFS
makes Build the same directory as build
test that a junk directory stays selectable while its contents stay out
of search results
* server: report a directory the walk could not read
a directory that fails to open or to iterate was skipped in silence, so
a caller got a listing that looked complete while a whole subtree was
missing: a path over the platform limit, a volume going away, a name the
filesystem rejects
skip_permission_denied never reaches this path, so an error here is an
incomplete answer rather than a deliberate omission, and it now sets the
truncated flag
* server: simplify the file_glob_search listing plumbing
return a small result struct instead of two out params and a caller path
that only fed an error string, taking list_entries from six parameters
down to three
scope the error code to the directory being read, act on the status code
the entry lookups already returned, and treat an unreadable link state as
a link so the walk never descends on a guess
check the deadline when a directory is popped, not only per entry, so a
tree of empty directories cannot outlive the budget
read the path parameter once, and reject an invalid limit the way an
invalid type is already rejected, instead of silently falling back
normalize the resolved path, so a "." or ".." a caller typed reaches
neither git nor the client, and return the generic path form with '/'
separators on every platform, so the base sent to clients no longer needs
a local fixup
* ui: expire cached picker searches
the cache grew for the lifetime of the component: entries went stale
after the TTL but were never removed, so every distinct query typed in a
session stayed in memory
drop expired entries when a new result is stored
* server: address review from @ngxson
trim comments to one line each, and drop two that restate the code
rename junk_lookup_name to get_effective_name, and move it and the link
check to private static members next to junk_dir_names
merge the Windows and Linux link checks into one is_link, so symlinks are
checked everywhere and junctions only add to it on Windows
* server: convert tool paths as UTF-8 on Windows
a narrow path uses the active code page there, so a file name came back
mangled and a path with an accent could not be opened at all
convert explicitly at every crossing between a std::string, which always
carries UTF-8 here, and fs::path
read the home directory through the wide environment, since the narrow
one returns the profile path in the active code page too
the walker no longer normalizes separators by hand, since paths now come
back in generic form
* server: fold the platform branch inside console_output_to_utf8
match the shape of the other helpers, one definition with the #if inside,
instead of two definitions wrapped in #if and #else
inline the single caller helper and trim the comment
a child process writes in the OEM code page, which is not UTF-8 on a
western Windows install, so accented output reaches the JSON layer as
invalid bytes and gets replaced there, silently losing the characters
run() spawns without a console, so the child never inherits the console
code page and GetOEMCP is the one that applies
decode with MB_ERR_INVALID_CHARS so a wrong code page returns the text
untouched instead of emitting replacement characters, and pass text that
already decodes as UTF-8 through so a child emitting UTF-8 is never
decoded twice
the check drops an incomplete trailing sequence before validating, since
a streamed chunk can end in the middle of a multi-byte character
* tests: add model resolution test on synthetic repo listings
Include download.cpp and arg.cpp inside a namespace with hf_cache
monkey patched to serve hardcoded listings, so the resolution and the
model handler assembly are tested end-to-end through the real CLI
parsing, without modifying the tested code and without network access.
Covers the primary, shard, mmproj, sidecar and preset resolution on
layouts mimicking real vendor conventions, replays every case on
permutations of the listing to assert determinism, and asserts the
final wired paths for the spec type auto-selection, the -md precedence
and the fallback suppression.
* tests: keep model resolution checks active and let the handler resolve
Replace assert with a REQUIRE macro alive in Release builds, key the
fake hf_cache by repo id so the real handler init resolves every plan
itself, check the exact shard sets, restrict the permutation exception
to the order dependent picks, and cover dflash and eagle3.
* tests: fix model resolution build on fatal warnings CI and Windows
The namespaced copy of the sources leaves many static functions unused
in this TU, exempt it from the unused warnings. Pre-include the
windows headers so arg.cpp does not pull them inside the namespace.
Declare the renamed copies of the download.h functions, verbatim from
the header and renamed in sync by the macros, so missing declaration
and missing prototype warnings are satisfied on every toolchain.
* tests: fix winsock inclusion order for the model resolution test
WIN32_LEAN_AND_MEAN and winsock2.h before windows.h, so http.h does
not redefine the socket types afterwards.
* tests: link cpp-httplib to the model resolution test
The test compiles its own copy of download.cpp, which calls httplib
directly, and the private link of llama-common does not propagate the
symbols under lld-link.
* common_http_client
* common: finish the http client wrapper
Add the virtual Head, Get and Post methods and the passthrough
setters to the common_http_client skeleton, move follow_location
into the constructor, expose the underlying client for the ranged
pull path, and rename the missed common_http_client_init call sites.
* tests: rewrite model resolution on the http client stub
Replace the namespace inclusion of the sources by a plain TU: the
common_http_client factory returns a stub serving hardcoded HF API
responses, so the real hf_cache parsing, resolution and CLI handler
run against synthetic listings in an isolated cache directory.
Failures print the named case, the reordering and the actual versus
expected values, the assembly cases use the full command line as
context, and the empty result cases are checked once to keep the
logs short.
* tests: fix the model resolution on Windows and the builds without TLS
Assert the exact expected paths composed like the cache does instead of
suffix matching on forward slashes, set the environment portably, and
serve the stub through an http endpoint so the builds rejecting the
https scheme still reach it. Pause the log so the negative cases can be
replayed on every reordering.
* tests: make the model resolution failures self explanatory
Resume the paused log before the failure report so the CI shows why
the tested code bailed, and format the stub oids portably.
* common: hold the http client factory behind exported functions
The factory was an inline variable, and the Windows shared builds
export functions but not data, so the executable and the DLL each had
their own instance: the stub installed by the test was invisible to
the library, which reached for the real endpoint and resolved nothing.
Route the creation through functions compiled into the library and
format the stub oids portably.
* common: add the http client factory source missed in the previous commit
* common: typedef the http client factory callback
Address review from @ngxson
* tests: serve the model resolution repos over the loopback
Replace the client stub by a real httplib server bound to the
loopback, so no C++ object crosses the module boundary anymore and
the library exercises its own client and transport end to end. The
debug shared build on Windows crashed inside the stubbed path.
* common: add portable common_get_env and common_set_env helpers
Address review from @ngxson
* common: drop the http client factory left without a caller
The loopback server made the stub substitution unnecessary, the client
init builds the real client directly again.
* common: read the model endpoint through the env helpers
* nit: drop the stub leftovers from the model resolution test
* common: align common_set_env and isolate the test cache per run
The POSIX branch now behaves like _putenv_s, so the helper has a single
contract on every platform, and common_get_env already reads an unset and
an empty variable alike.
The model resolution test keys its cache directory on the loopback port,
where two concurrent runs on the same machine used to share one directory
and the initial cleanup of either wiped the other.
* tests: move the model resolution server into main
* tests: support the DSpark sidecar resolution
* common: revert the http client to the plain httplib client
address review from @ngxson
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
The dspark- files resolve like the other speculative sidecars: the
-hfd tag applies to them, a requested sidecar resolves without a full
model at the tag, and an explicit -md selection disables the discovery.
When no type is requested, dspark outranks dflash in the auto-selection
since its sidecar carries the extra Markov head.
* common: fix explicit -md precedence over draft sidecar resolution
Follow-up of #25955, an explicit --model-draft file given with -hfd
was silently overridden by the sidecar resolution of the draft repo,
and its path was never resolved to a local file.
An explicit draft file selection now disables the sidecar resolution,
so the manual CLI configuration wins over the automatic one.
* common: apply the -hfd tag to the sidecar resolution
The sidecar selection was anchored on the primary of the draft plan,
so a tag without a matching full model aborted the whole plan, and
the sidecar quant silently followed the default model pick.
The tag now anchors the sidecar directly: exact tag match first, then
closest quant to the tag, and a requested sidecar resolves even when
no full model matches the tag. A wired draft sidecar also counts as
an explicit draft, so the main plan no longer downloads a second one.
* common: promote speculative load logs from trace to info
Show the loaded draft model and the MTP draft context at the default
verbosity, for consistency with the mmproj and primary logs.
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* server + ui: refactor resumable stream routes to query string conv_id
The conversation id can embed a model name containing slashes
(ggml-org/...) in router mode, which the decoded path splits before the
:conv_id param is captured, so stop and resume never matched the
session. Move the id to the conv_id query string on the public routes
and on the internal router -> child hop, where slashes survive
encoding. Handlers are unchanged since query and path params land in
the same map. Add a regression test with a slashed model name.
* server: move stream route docs to server-stream.h
Address review: ngxson wants the main server.cpp registration code kept
clean and simple, with route-level explanations living in the header.
Move the query string rationale and the lookup ownership note next to
the handler declarations in server-stream.h, and shorten the wiring
comment to a pointer.
* server: cancel a pending request when its stream is stopped during model load
The conversation was registered in the conv map only after the blocking
autoload wait, so a stop issued while the model loaded found nothing to
cancel and the request went on to generate an orphan once the load
ended. Register the conversation before the wait and give the entry a
ticket: a stop erases the entry, and the parked request checks its
ticket after the wait and aborts with 400 instead of starting. A newer
request on the same conversation replaces the entry, so only the
stopped request is cancelled. Add a regression test that stops during
the load window.
* server + ui: resume a stream after a page reload during model load
A pending request died with the client socket when the page was
reloaded while its model was loading, so no session ever existed and
the conversation had nothing to recover. A session request that waited
for a load now detaches from the client socket and reaches the child
regardless, the session buffer receives the generation, and the resume
route answers 503 while the owner is loading so the client retries
instead of dropping its state. The WebUI persists the pending stream at
send time, quietly polls on 503, and attaches once the session exists.
Add a regression test that drops the client during the load window.
* ui: show the model load progress again after a page refresh
The resume wait was invisible, so a conversation refreshed while its
model was loading showed nothing until the first byte. On a 503 from
the resume probe, mark the conversation as loading again so the
assistant row persisted at send time renders the processing info, and
target the model frozen in the persisted stream state for the
progress, since the row has no model yet and the dropdown may not be
restored.
* fix CI
* fix CI bis
* ui: detect the conversation import format from file contents
iOS resolves every accept entry to a UTI and has none for ".jsonl", so the
picker greyed out exported conversations. Drop the accept filter and pick the
parser from the file contents: ZIP magic bytes, then a first "session" record
for JSONL, otherwise the legacy JSON format.
Also remove the unused importConversations() picker and an orphan doc comment,
and cover each format with unit tests.
* ui: report what a conversation import actually wrote
The import summary echoed the selection back, so re-importing conversations
already in the database claimed success while nothing was written and only a
console warning said otherwise.
Return the imported and skipped conversations from the database layer, list the
written ones in the summary, and count the rest in a toast.
* ui: name the literals of the JSONL conversation format
Introduce SessionRecordType and SESSION_HARNESS, and reuse the existing
NEWLINE constant, so the record format lives in one place.
This also covers the writer side, which predates the import path under
review and carried the same literals: an enum stated by the reader alone
lets the two sides drift.
Values are unchanged, so an export stays byte identical.
The context gauge card starts monitoring like the dial does, because
its own processing state instance only follows the live stream while
its monitoring flag is set. It also gets back the text-sm and ring
classes the removed hover card wrapper used to inject, which restores
its layout. Routing to a conversation now lands at the bottom
instantly and keeps the pin one frame at a time until the page height
settles, since content-visibility size realizations and syntax
highlight passes grow the page without DOM mutations.
* ui: remove viewport fade in and smooth autoscroll bottom snap
fadeInView mounted every message and markdown block at opacity 0 and
relied on an IntersectionObserver to reveal it. When the observer never
fires (blocks mounted offscreen during long agentic loops) the content
stays invisible forever while still present in the DOM. Remove the
action, its orphaned isElementInViewport util and all call sites:
blocks now render visible immediately.
AutoScrollController.scrollToBottom defaulted to behavior smooth and is
invoked every 100 ms while streaming. Each tick restarts an easing
animation toward a moving scrollHeight, producing a random elastic bump
of a few pixels when the user reaches the bottom and autoscroll
reengages. Default to instant scrolling; the user facing scroll down
button keeps its smooth behavior.
* ui: skip rendering of offscreen chat messages via content-visibility
Apply content-visibility auto with contain-intrinsic-size to chat
messages so the browser skips layout and paint for messages outside
the viewport. The DOM stays complete: component state, find-in-page,
text selection, and the mutation based autoscroll are unaffected, and
browsers without support simply ignore the properties.
* ui: remove conversation switch fade
Switching conversations faded the message list out and in over 500 ms
plus a 300 ms route delay, deferring the message refresh behind two
requestAnimationFrame calls. Remove the fade, its navigation hooks and
dead state, and refresh messages directly so switching is only bound
by actual render time.
* ui: describe present behavior in comments and drop unused parameter
* ui: anchor the context gauge popup to the form with plain CSS
The stats card was portaled to body and repositioned in script on
every ancestor scroll event, trailing the page by one frame while
streaming. Render it as an absolutely positioned sibling of the input
box inside the already relative form, so nothing runs during scroll.
The card sits just above the dial, centered on it and overlapping the
textarea, from a single measurement of the dial center and top taken
when it opens; the dial and the card share the same positioning frame,
so the values stay exact while the card is open. Mouse pointers open
on hover with a grace delay to reach the card, touch pointers toggle
on tap, any press outside the card and the dial closes it, and Enter
and Space toggle from the keyboard. The card lives outside the input
box because its overflow-hidden and backdrop-filter would clip any
positioned descendant.
* ui: extract context gauge popup constants and relocate its state store
Move the placement values and the close grace delay to
lib/constants/context-gauge-popup.ts, matching the auto-scroll
constants layout, and move the popup state module from the component
folder to lib/stores where runes modules live in this codebase.
* ui: declare the context gauge popup card ref as $state
* ui: fix MCP server display name conflicts in tools lists
Tool groups were keyed by display label so two servers reporting
the same name broke the keyed each blocks and only one was visible.
Key rendering, expand state and toggles by the stable server id
instead, and suffix duplicate labels with a counter in config order.
* ui: customizable MCP server display name with autofill
Add a display name field to the MCP server form, add and edit alike.
The custom name takes precedence over the server-reported one, so two
servers reporting the same name can be told apart; clearing the field
returns to the automatic label. In the add dialog a debounced preview
handshake prefills the field with the server-reported name: a manual
edit freezes the autofill, stale responses are discarded, failures
stay silent, and an unedited prefill is not persisted so the label
keeps following the server.
* ui: fix recursive fetch passthrough in the client test setup
The original fetch was captured inside beforeEach, where it is the
previous test's spy since vi.spyOn returns the existing one, so the
default passthrough recursed on itself for any URL outside the
mocked set. Capture the real fetch once at module load.
* ui: add Default reasoning option that defers to the server
The webui always injected enable_thinking, overriding the chat template
default and the --reasoning flag, breaking models that reason
unconditionally (e.g. Gemma 4 E4B) on a fresh client.
Default sends nothing so the server decides, Off and effort levels
force the value as before. All choices are remembered.
Also remove the boolean thinking API from the conversations store and
drop ChatFormReasoningEffortSubmenu.svelte (dead code).
* ui: close the whole menu tree on reasoning level selection
The reasoning levels were raw buttons inside the SubContent, so
selecting one only closed the submenu via manual state while the root
dropdown stayed open. DropdownMenu.Item closes the full tree on select
like the sibling entries and brings native keyboard navigation.
* ui: prevent the add menu tooltip from flashing when the dropdown closes
With -hfd pointing to a repo that ships mtp-/dflash-/eagle3- sidecars
and no --spec-type given, the draft resolved to a full model while the
sidecar was the intended draft.
When the speculative types are still at their default, discover the
sidecars of the draft repo, pick the first available following the
existing mtp > dflash > eagle3 priority, and set the corresponding
type, so this now works without any extra flag:
llama-server -hf repo:Q3_K_M -hfd repo:Q8_0
An explicit --spec-type disables the inference, and a draft repo
without sidecars keeps resolving to a full model as before.
* cuda: add k-quant support to GET_ROWS
Device-side embedding lookups require GET_ROWS to handle the k-quants
used by common GGUF recipes (Q4_K_M stores token_embd as q6_K). Without
it the backend rejects the op and the scheduler falls back to the host,
copying the full embedding matrix back on every token in single-device
graphs.
Factor the super-block dequantizers out of the dequantize_block kernels
in convert.cu into shared device functions in dequantize.cuh and reuse
them from a new k_get_rows_kq kernel : one thread block dequantizes one
(dst row, super-block) pair with the existing thread layouts, 32 threads
for q4_K and 64 for the other k-quants.
Covers q2_K to q6_K in get_rows_cuda and supports_op. i-quants are left
as a TODO.
* cuda: add i-quant support to GET_ROWS
Extends the shared super-block dequantizers to the nine i-quants and
reuses them from k_get_rows_kq with the 32-thread layout of the matching
convert.cu kernels. supports_op gates the k-quant and i-quant path on
ne0 being a multiple of QK_K, which iq4_nl does not guarantee on its
own (QK4_NL sub-blocks). mxfp4 is left as a TODO.
* cuda: add mxfp4 support to GET_ROWS
Moves the mxfp4 dequantizer into the shared super-block helpers and
reuses it from k_get_rows_kq with the 32-thread layout of the matching
convert.cu kernel. mxfp4 joins the ne0 % QK_K gate in supports_op since
its 32-value sub-blocks do not guarantee QK_K-aligned rows on their own.
This closes GET_ROWS type coverage on CUDA: every quantized GGML type
now takes the direct device path.
* cuda: gate the GET_ROWS row size only for 32-value sub-block types
Address review from @pwilkin: the i-quant commit replaced the return
shared by the whole supported type cascade, so f16/f32/bf16/i32 and the
legacy quants also inherited the ne0 % QK_K == 0 gate and any row size
that is not a multiple of 256 fell back to the scheduler. Split the
cascade: unconditional support is restored everywhere, the gate stays
only on iq4_nl and mxfp4 whose 32-value sub-blocks do not guarantee the
QK_K super-blocks the kernel iterates on.
With -hfd pointing to a repo shipping speculative sidecars, the draft
resolved to the main model of that repo, since find_best_model()
excludes sidecar files, and the explicit draft plan suppressed the
sidecar discovery on the -hf repo.
The draft plan already discovers its sidecars, they were just never
consumed. Wire them as the draft, following the fallback pattern of
the main plan, so this now works as expected:
llama-server -hf repo -hfd repo --spec-type draft-dflash
* server: return 400 instead of 500 on validation error with X-Conversation-Id
set_req() attaches the spipe as soon as the header is present, before the request
body is parsed. When params validation throws, set_next() never runs and next_orig
stays empty, so on_complete() called it and crashed with std::bad_function_call,
turning the prepared 400 JSON into a generic 500.
on_complete() now treats an empty next_orig as "streaming never started" and evicts
the session installed by set_req(), so a failed request leaves nothing behind for
discovery or replay. This also covers valid requests that carry the header but do
not stream, which previously left an empty finalized session in the map until the
GC TTL.
* ui: do not send the backend_sampling placeholder
On a fresh profile the syncable settings hold the empty string placeholder meaning
"let the server decide". Every neighbor field goes through the hasValue() guard
that filters it, except backend_sampling, which sent the placeholder verbatim and
made every default settings completion fail validation.
Guard the field with hasValue() like its neighbors. hasValue(false) is true, so an
explicit false still reaches the server and the intent of #18781 (send both true
and false) is preserved. Only the placeholder is filtered.
Edge paragraph margins are now zeroed at the source in
markdown-content.css, but the user bubble and the system message still
carried the -my-4 compensation for them. The uncompensated negative
margins shrank the wrapper 2rem below its content, collapsing
single-line user bubbles into a scrollable sliver and skewing the
system message expand threshold.
* ui: fix Show tool call in progress toggle ignored
The showToolCallInProgress setting was disconnected from the render
path during the agentic content rework: getDefaultExpanded() returns
a hardcoded false for tool call sections and an unconditional effect
auto-expands the currently executing tool call regardless of the
setting.
Drive default expansion of all tool call section types from the
setting and remove the now redundant auto-expand effect. Manual
toggling still takes precedence over the default in both directions.
* ui: rename Show tool call in progress to Always show tool call content
The previous name suggested symmetry with Show thought in progress,
which only applies while inference is running, but tool call content
stays expanded after completion. Rename the label, the settings key
and the syncable server key to alwaysShowToolCallContent. The synced
parameter never worked under its previous name so no migration is
needed.
The agentic gate counted MCP servers, builtin and custom tools but not
frontend tools, so with the JS sandbox as the only tool source the
agentic flow was skipped, no tools field reached the server and the
chat template rendered without the tool system prompt.
The sandbox is fully client-side: frontendTools derives from the
Developer settings toggle alone, counting it in the gate restores that
single source of truth.
read_file with append_loc emits "{n}\u2192 {line}". The space after the
arrow is meant as a separator, but it is indistinguishable from real
indentation. Models strip "{n}\u2192" yet keep the space, so the old_text
passed to edit_file carries a phantom leading space and never matches
(normalize_for_fuzzy_match trims trailing whitespace only, never leading).
Drop the separator space so the arrow abuts content: stripping "{n}\u2192"
now yields the exact line with its real indentation preserved, and the
failure mode cannot occur by construction. Update the description example
to match the new format.
In MODEL mode, modelPropsCache is never populated: fetchModelProps
call sites are gated on router-only state (isRouterMode checks,
routerModels always empty), so supportsThinking always reads an
empty chat template once a model is auto-selected.
Read serverStore.props.chat_template directly in non-router mode,
since the global /props already describes the single loaded model.
* metal: fuse snake activation (mul, sin, sqr, mul, add)
Mirror the CUDA, Vulkan and CPU snake fusion: same matcher on the naive
5-op chain, same F32 contract on a and inv_b, same F32/F16/BF16 kernel
with F32 compute. Follows the Metal backend idioms: bf16 instantiation
gated behind GGML_METAL_HAS_BF16 and concurrency ranges checked on the
remaining chain nodes before encoding, as done by the bin fusion.
Covered by the existing backend-agnostic SNAKE_FUSE tests.
* metal: absorb snake fusion into ggml_metal_op_bin
Extract the matcher to ggml_metal_op_can_fuse_snake, mirroring the
Vulkan naming, and dispatch the fused path from ggml_metal_op_bin.
The encode loop switch is back to a single call per case.
Address review from ggerganov
* metal: fix indentation in ggml_metal_op_can_fuse_snake
* ui: fix MCP panel regressions after settings rework
Restore the llama-server proxy switch in the Add New Server dialog.
The dialog never passed useProxy/onUseProxyChange to McpServerForm,
which only renders the proxy switch when the handler is provided.
The flag is now wired, persisted on addServer, and reset on close.
Bound the MCP connection handshake with the configured timeout.
handshakeTimeoutMs was set in the server config but never consumed.
The SDK timeout only covers the initialize request, not
transport.start(), which can hang forever on an unreachable host.
The whole handshake now races against the timeout and closes the
transport on expiry so the underlying fetch or socket is aborted.
Keep disabled MCP servers visible in management and chat-add UIs.
Collapsing mcpDefaultServerOverrides into mcpServers[i].enabled turned
the visibleMcpServers enabled filter into a visibility trap: toggling
a server off outside a conversation hid it from every surface with no
way to re-enable it. The filter is dropped, tools derived from health
checks still skip disabled servers, and the settings page and server
card render the real card instead of a skeleton for disabled servers
that never receive a startup health check.
* ui: clarify MCP server list semantics and add regression test
Remove the visibleMcpServers getter, a filterless alias of getServers
whose name invites the next refactor to put a filter back. Call sites
read getServers directly, the duplicate list in the chat submenu is
merged, and the misleading local variable in the sheet is renamed.
A parser unit test pins the invariant: enabled is an on/off state,
never a visibility filter, so disabled servers stay listed and
toggleable.
* ui: apply the MCP request timeout setting live to all servers
The per-server requestTimeoutSeconds field was never editable in any
UI and froze the global setting at server creation time, so changing
the timeout in Settings was a no-op for existing servers. The field
is removed from the data model and parsers, the timeout is read live
from the global setting wherever a request config is built, and the
misleading "Can be overridden per server" help text is dropped. A
parser unit test guards against reintroducing the stored field.
* ui: move the MCP request timeout into the Agentic settings section
The MCP section held a single setting. The timeout is a global tool
execution parameter like the other Agentic entries, so it moves there
and the section is removed. Same settings key, no migration needed.
* ui: remove the dead tool preview lines setting
The agenticMaxToolPreviewLines setting was read into AgenticConfig
and consumed by nothing: the agentic loop only uses enabled and
maxTurns. Its help text described a previous architecture where only
truncated previews and the final response survived the loop; tool
results and intermediate turns now persist as full DB messages, so
the setting had no effect at any value. Stale keys in localStorage
or a server ui-config are ignored.
* ui: resolve absent MCP per-chat overrides to the server enabled flag
New conversations started with every MCP server off: the settings
rework stopped seeding a per-conversation override list, assuming
the enabled check would fall back to mcpServers[i].enabled, but it
fell back to false, and the send path passed the raw stored list
with no fallback at all. The per-conversation list is now sparse by
contract, holding only explicit toggles, and every access point
resolves a missing entry to the server's own enabled flag: the
toggle display, the resolved list handed to the agentic flow, and
the enabled check itself.
The mobile "+" sheet was missing the reasoning effort section present
in the desktop dropdown, so thinking could not be toggled on touch.
Extract the shared derivation and selection logic into useReasoningMenu
and consume it from both the desktop submenu and the mobile sheet,
keeping a single source of truth and preserving each surface idiom.
* mtmd: fix silent prompt truncation on embedded NUL
mtmd_input_text carried the prompt as a bare const char* with no
length, so a NUL byte in message content cut the prompt at the
tokenizer boundary and dropped every later message plus the assistant
marker, with no log. Add an explicit text_len and thread it through,
matching llama_tokenize and the text only path.
* cleanup
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* server: accept null sampling params
Extend the schema validation to treat a null value as absent, so
clients can send null on nullable params (temperature, top_p, ...)
to request the server default. This matches the OpenAI spec and the
json_value convention used elsewhere.
Add has_field() to skip null in the field eval guards.
* has_field -> has_value
downloadConversation serialized activeMessages, the root -> currNode
path, so exporting a conversation with edited or regenerated messages
dropped every alternate version and kept only the selected one.
Fetch the whole message tree via getConversationMessages so the export
carries all message versions, matching the multi-conversation export
path which already did this. Keep the active conversation as the header
source to preserve an up-to-date currNode.
Forks are separate conversations, each with its own convId, and are
exported on their own.
* cuda: fix snake fusion type predicate, a and inv_b are F32
The matcher required a->type == x->type while launch_snake reads both
as const float *, matching the CPU and Metal contract where a and inv_b
stay F32. F16/BF16 chains never fused and fell back to the naive path,
and a hypothetical all F16 chain would have read F16 bits as float.
Aligns the predicate and the comment with ggml-cpu.c
* cuda: reject snake fusion on non-contiguous operands
The kernel reads x[idx] and a[c] / inv_b[c] linearly, so a
non-contiguous view passing the matcher would silently read wrong data.
Mirror the contiguity guard already present in the CPU, Vulkan and
Metal matchers.
* server-stream : pimpl
* server-stream: prefix free functions with server_stream_
address review from ggerganov: scope the public stream functions under the
server_stream_ prefix, matching server_stream_session_manager_start/stop.
* server-stream: guard session and manager state with the mutex
address review from ggerganov: make done, completed_ts and the GC running flag plain members under their
mutex and set the condvar predicates under the lock. keep cancelled atomic for
the lock-free should_stop poll.
* server-stream: trim comments to the non-obvious
address review from ggerganov: drop comments that restate the code, keep the
concurrency, lifetime and ordering rationale. de-stale a few comments left by the
pimpl: g_stream_sessions is now internal and the /v1/streams listing is gone.
* server-stream: update dev docs for the pimpl and prefix
reflect server_stream_session_manager_start/stop and the server_stream_ prefix,
note the manager is now a file-static singleton hidden in the .cpp
* server-stream: move stream traces to debug level
keep the bring-up traces for diagnostics but off the default log: skip
drain, draining, drain ended, DELETE evict, attach_pipe, and the router
stream resume proxy.
* server-stream: align router stream resume proxy trace with upstream
the child-side bring-up traces are already SRV_TRC on master, move the
router stream resume proxy trace to the same level.
* server-stream: move stream_read_status enum to the cpp
it is only used by the hidden session and consumer types, so it belongs
with them behind the pimpl boundary, not on the public header surface.
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* metal: add col2im_1d op (f32/f16/bf16)
Gather kernel mirroring the CPU/CUDA path: each output (t_out, oc)
reads its ceil(K/s0) source columns with an F32 accumulator, a single
write and no atomics. One thread per output element, 256 per
threadgroup.
* metal: check dst contiguity and type match in supports_op for COL2IM_1D
Align the GGML_OP_COL2IM_1D predicate with the CPU, CUDA, and Vulkan
backends: the kernel writes dst with linear indexing and assumes the
same type as src0, so supports_op must also require a contiguous dst
and op->type == op->src[0]->type.
* Update ggml/src/ggml-metal/ggml-metal.metal
Co-authored-by: YiChen Lv <63285796+forforever73@users.noreply.github.com>
---------
Co-authored-by: YiChen Lv <63285796+forforever73@users.noreply.github.com>
* server: fix deadlock in load_models() when erasing a finished download
The download monitoring thread acquires the models mutex on its way out,
but load_models() joined it from the erase loop while holding that mutex.
Join it outside the lock via threads_to_join like the other monitoring
threads.
* server: add default timeout to test requests
A hung server now fails the test after 10 minutes instead of stalling
the CI job for hours. Explicit timeouts are unchanged.
* ui: migrate legacy string-encoded booleans in persisted config
* ui: enable thinking by default
Fresh users and legacy conversations without a persisted thinking
preference now default to enabled. The per-conversation toggle and
the persisted localStorage choice keep taking precedence.
Picks up the enable_thinking default from #24876.
* server + ui: ping silent SSE streams every 1s and kick only after 3s so slow prefill never drops healthy connections
* server + ui: sse_ping_interval becomes a per-request body field
Address review from ngxson: the global default returns to 30 so API
clients see no behavior change, and the WebUI sends sse_ping_interval: 1
in the request body since it owns the 3s visibility-kick contract and
declares the cadence it needs. Positive values keep the existing > 0
gate, -1 keeps its disabled semantics.
* server: move sse_ping_interval into the request schema
Address review from ngxson: the field is now a typed field_num with
hard limits (-1, INT32_MAX) bound to task_params, seeded from the CLI
default alongside the other inherited parameters. The raw json_value
read and its redundant comment are gone, and schema evaluation brings
type and range validation for free.
* common,server: handle bracketed IPv6 literals in URL authority
Parse the [host]:port form (RFC 3986) and bracket IPv6 hosts when
formatting a URL authority: listening log, proxy Host header, proxy
log, client rebuild. The per-request remote_addr stays bare.
* common: restore unsupported scheme throw in url parser
Address @ngxson review: keep the explicit reject in port resolution so
the block stays self-contained. Non-http(s) schemes still throw (also
gated at the top of common_http_parse_url).
Pascal
