* llama : add GLM5-Next NextN (MTP) graph
Build the GLM5-Next multi-token-prediction head as graph_mtp: the NextN block
embeds enorm(tok)+hnorm(h) through eh_proj, runs one plain DSA layer and the
shared lm_head, reusing the trunk's builders through the no_build tag ctor.
llama_memory_recurrent also tolerates a partial seq_rm when the context holds
no recurrent layers, which is what the MTP draft context needs.
Assisted-by: Claude
* llama : glm5-next: skip dead compute in headless NextN forwards
A NextN forward with no output rows (the MTP catch-up and the draft-context
prefill) persists only through its cache writes, so the headless graph keeps
the MLA latent, indexer key|gate and pooled-key writes and drops the query
path, the indexer selection, the attention body, the FFN and the LM head. The
4-token catch-up falls from 6.9 ms to 0.33 ms of kernels; the greedy output
hashes and the draft acceptance are unchanged.
Assisted-by: Claude
* llama : glm5-next: fix NextN extraction contracts and shared-tail rollback
Three fixes from the architectural review. The headless graph prune now also
requires that no unmasked nextn extraction is live, because that mode reads
n_tokens hidden rows regardless of the logits flags. Masked extraction
publishes the hidden rows gathered by the output ids, so a batch whose output
flags are not a prefix exports the right rows. A partial recurrent rollback
whose tail cell is shared with another sequence is now rejected instead of
silently moving that sequence's tail.
Assisted-by: Claude
* llama : glm5-next: tidy comments in the MTP changes
Assisted-by: Claude
* llama : glm5-next: crop the MTP graph to the output rows instead of pruning it
Replace the headless NextN prune with the crop pattern the other MTP
graphs use: gather the attention output and the block input at the
output ids before the position-wise FFN and the shared head. A NextN
forward with no output rows (the MTP catch-up and the draft-context
prefill) then runs the FFN and the head over zero rows. The 4-token
catch-up falls from 6.9 ms to 2.9 ms of kernels; greedy output hashes
are unchanged.
Assisted-by: Claude
* glm5-next: use the nextn crop helpers in the MTP graph
Replace the local crop condition and the masked select of t_h_nextn
with crop_before_nextn and crop_after_nextn, so the MTP graph narrows
its rows the same way as the main graph and the other models.
Describe the shared cell and empty filter branches of the recurrent
partial rollback.
* glm5-next: load MTP-only and trunk-only GGUF files
Make the trunk tensors optional when the file only holds the NextN
layer, and the NextN tensors optional when the file only holds the
trunk, so the split MTP GGUF loads as a draft model.
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Pascal <admin@serveurperso.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
The speculative MTP init enables NextN extraction on the target and draft
contexts after both were created and their schedulers reserved. With
unmasked extraction the trunk graph keeps every token through the last
layer instead of cropping to the output rows, so the first decode
reallocates to that batch's shape and the next, wider batch trips
GGML_SCHED_DEBUG_REALLOC. Invalidate the reserve when the flags change so
the next compute re-reserves with the new graph shape.
Assisted-by: Claude
* vulkan: optimize IQ4_XS matmul kernels
Assisted-by: OpenAI Codex
* vulkan: address IQ4_XS review nits
- drop the dead LOAD_VEC_A != 8 branch in the IQ4_XS shmem load; iq4_xs is
in lut_load_vec_a()'s "8" list, so that path is never generated
- disable MMVQ for IQ4_XS on Intel (27.3% tg regression on A770)
- remove a stray empty line in types.glsl
Assisted-By: Claude Opus 5 <noreply@anthropic.com>
convert_unary handles the contiguous case through the general strided kernel,
one element per thread: each lane reads 4 bytes and writes 2. Converting the
activations for a bf16 matrix multiplication that way moves 126 MB in 1021 us
on gfx1151, about 65% of what the memory system can do.
Give the contiguous path its own kernel that takes four elements per thread
through a vector type, so a warp loads 512 bytes at a time instead of 128. It
is used only when the element count is a multiple of four and both pointers
carry the alignment the vector type needs, and falls back to the strided
kernel otherwise.
Model level, Qwen3.8-Next-Flash IQ3_XXS on gfx1151, llama-bench -ub 2048 -r 6,
mean of the last 3 reps, ABBA counterbalanced:
pp2048 688.0 680.0 -> 694.3 691.1 +1.26%
tg128 24.8 24.8 -> 24.8 24.8 +0.14%
Every conversion in a prefill takes the new kernel (kernel trace: 1146
convert_unary_cont_vec4, no convert_unary). Output is bit identical; MUL_MAT,
MUL_MAT_ID, CPY, CONT, GET_ROWS and SET_ROWS pass.
Assisted-by: Claude Opus 5
* HIP: enable mma FA for head size 256 on RDNA4, tune configs
Assisted-by: Claude
Assisted-by: Codex
* HIP: prefer whole-tile FA grids over stream-k on AMD WMMA
Assisted-by: Claude
Assisted-by: Codex
* revise stream_k logic
* revise kernel selection logic
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* chat : split specialized parsers into common/parsers
Move the 14 dedicated template parsers out of chat.cpp into one file each under
common/parsers, mirroring the src/models split. chat.cpp keeps the template
detection in common_chat_try_specialized_template() and drops from 3915 to 1513
lines.
common/parsers/parsers.h holds the shared helpers and one declaration per
parser. foreach_function/foreach_parameter become inline there since nothing in
chat.cpp uses them any more; common_chat_template_direct_apply_impl and
common_chat_template_generation_prompt_impl lose static and carry their default
arguments in the header. Parser-specific helpers move with their parser:
is_lfm2_template, deepseek_v4_sort_tool_results and the gemma4 turn builder.
No functional change.
Assisted-by: Claude Opus 5
* chat : enumerate parser sources instead of globbing
file(GLOB) does not re-run CMake when a source file is added or removed, so an
incremental build silently keeps building the old set. List the parsers in
common/parsers/sources.cmake and include it from common/CMakeLists.txt.
Assisted-by: Claude Opus 5
* split helpers, add newlines
* model: add Kimi-K3 text model
Hybrid KDA (linear) + MLA (full) attention as in Kimi-Linear-48B, plus five
things that architecture does not have:
1. cross-layer residual attention (attn_res_block_size)
2. latent MoE (routed experts run at n_expert_latent)
3. situ activation (replaces SwiGLU everywhere)
4. MLA output gate (sigmoid gate before o_proj)
5. full-rank KDA gate (single ssm_g instead of ssm_g_a/ssm_g_b)
K3's text_config reports KimiLinearForCausalLM - the older 48B architecture -
so get_model_architecture routes on the top-level name instead.
The KDA decay gate has two forms, selected by linear_attn_config's
gate_lower_bound. It is not a clamp: when set it swaps the activation entirely
(fla/ops/kda/gate.py), from -exp(A_log)*softplus(x) to
lower_bound*sigmoid(exp(A_log)*x). K3 sets it to -5.0; kimi-linear leaves it
unset, so that path is unchanged.
Cross-layer residuals reuse ggml_dsv4_hc_pre for the weighted sum. That op is
CPU + CUDA only, so Metal/Vulkan will fall back per-node until those kernels
exist.
The routed experts ship as compressed-tensors "mxfp4-pack-quantized". That is
bit-compatible with ggml's MXFP4 - same E2M1 code assignment, same E8M0 scale
byte, only the nibble positions within a block differ - so they are repacked
rather than dequantized, losslessly and without a ~5.5 TB bf16 round-trip.
The repack is built lazily because gguf_writer holds every added tensor until
the final write. DeepSeek-V4 was already doing the identical bit-shuffling, so
it now shares the helper.
Verified against Moonshot's own code path (transformers + fla's Triton KDA
kernels) on a tiny model exercising every K3-specific feature. Final-position
logits vs the fp32 reference: 6.7e-05 rel / corr 1.00000000 for both the
chunked and the recurrent delta-net path. MXFP4 blocks dequantize to the source
weights with 0.0e+00 error.
Assisted-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* model: fix ty errors in the Kimi-K3 converter
- `_res_parts` buffers (kind, tensor) pairs, not bare tensors
- `get_tensors` must return an Iterator, matching ModelBase
- LazyBase's `func` takes one argument, so pass the expert loaders through
`args` instead of the closure
- borrowing KimiLinearModel.set_vocab from an unrelated TextModel is
deliberate and safe, but not expressible in the signature
No behaviour change: the MXFP4 repack still dequantizes to the source weights
with 0.0e+00 error and end-to-end logits are unchanged (8.386e-03 rel,
corr 0.99996630).
Assisted-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Update conversion/kimi_k3.py
Co-authored-by: Boris Dvorkin <b_dvorkin@niuitmo.ru>
* Increase LLAMA_MAX_EXPERTS from 512 to 1024
* tests : support for Kimi K3 in archs test
* chat : add Kimi K3 chat format (reasoning, content, typed tool calls)
K3's assistant output is an XTML-ish tagged format built by the template's
open_tag/close_tag macros. Two properties break generic parsing:
1. The generation prompt ends with open_tag('think'), so the completion
starts inside the think section with no opening marker in the output
(thinking_forced_open).
2. Only <|open|>/<|close|>/<|sep|>/<|end_of_msg|> are special tokens; tag
names ("think", "response", "message") are ordinary text tokens.
Adds common_chat_params_init_kimi_k3 (PEG_NATIVE) with detection on the
marker trio, reasoning extraction, response unwrapping, and tool-call
parsing of the tools/call/argument tag structure with argument types
taken from the tool schema. Includes the K3 chat template fixture and 9
test-chat cases derived from real generations of the full 2.8T model.
Verified end-to-end against Kimi-K3-Q2_K (GrEarl/Kimi-K3-GGUF) on 8x B200:
content, reasoning_content, streaming deltas, and tool_calls all correct;
finish_reason stop/tool_calls as appropriate.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* chat : add message_delimiters for Kimi K3
Per-role message-start markers for token-level span splitting. User and
assistant messages carry only the role attribute, so their full opener
(through <|sep|>) is used; system and tool messages continue with more
attributes (type=/tool=/index=), so those delimiters stop after the
role's closing quote. Verified against the K3 tiktoken vocabulary that
the closing quote is always a standalone token across all attribute
variants, so the token-level prefix match stays exact.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix: apply nits from @ngxson and text fixes from @danielhanchen
* tests : added missing hyperparameters and tensors for Kimi K3 in test-llama-archs
* chore : move overly verbose header file comments to Kimi K3 source file
* tests : re-enabled KIMI_K3 in test-llama-archs for WebGPU backend
* model-saver : emit kda_gate_lower_bound for Kimi K3
Quick fix. The Kimi K3 loader reads kda_gate_lower_bound and gates a graph branch on it (it scales the KDA gate when the bound is above -INFINITY), but the model
saver never wrote the key, so a save->load roundtrip silently dropped it back to the -INFINITY default and changed the model's output. The real K3 config sets gate_lower_bound = -5.0.
I propose to emit it from the saver, and set it to -5.0 in the test-llama-archs K3 case so the roundtrip check exercises it (the roundtrip fails without the saver line).
* Refactor conditional for model architecture check
* tests : re-enabled (again) KIMI_K3 and MINIMAX_M3 in test-llama-archs for WebGPU backend
* fix code comments
* add template on conversion
* move repack_mxfp4_blocks to model base
* nits
* add_value_length
* optimize res_stack construction
* nits
---------
Co-authored-by: Boris Dvorkin <b_dvorkin@niuitmo.ru>
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
Co-authored-by: Deepankar Singh <singh.deepankar39@gmail.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Caleb DeLeeuw <caleb.deleeuw@gmail.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* chat: fix DS4 template to explicitly follow reference behavior
* Support DeepSeekv4 flag (`drop_reasoning`).
* fix: hook DS3.2 parser for DS4 as well
* fix: add tool result reordering
* fix: post-merge
k_get_rows_float did a scalar one-element-per-thread copy and recomputed the
row-invariant work (index load, fast_div_modulo, src/dst row pointers) for
every element. Hoist that out of the per-element loop, and add a vectorized
path (k_get_rows_float_vec) that copies one int4 (16 B) per thread for the
contiguous same-type (no-cast) case.
The vectorized path is gated at compile time (is_same<src0_t, dst_t>) and at
runtime on 16-byte alignment of the base pointers and all row strides and on
ne00 % VEC == 0. Vectorizing divides the block count by VEC, so a small
single-row gather can drop below the device CU count and regress; an
occupancy gate keeps those on the block-rich scalar path.
On Strix Halo (gfx1151) the DeltaNet recurrent-state gather (ne00=524288)
drops 18.6us -> 13.0us (rocprofv3 HW timestamps), faster than the Vulkan
backend, with no regression on the small conv-state gather; total get_rows
-27%. test-backend-ops GET_ROWS passes (47/47).
Assisted-by: Claude Opus 4.8
* meta: add hard emphasis on agents not writing descriptions/comments
Add a block in AGENTS.md to emphasize that agents are forbidden, under any circumstances, to post comments or pull request descriptions on behalf of the user.
* Add example
* Move examples to examples
* White space
* chat: trim messages sent to StepFun parser (fixes long reasoning loops)
* add regression test; remove duplicate template
* chat: trim StepFun content parts before rendering
The StepFun trim workaround ran on the already-rendered messages, where
typed content parts have been concatenated into a single string, so the
per-part whitespace could no longer be reached. Move the trim ahead of
rendering and apply it to content_parts text as well as the string
content and reasoning_content. Adds a content-parts regression test.
Co-Authored-By: Piotr Wilkin <ilintar@gmail.com>
Assisted-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: tarruda <tpadilha84@gmail.com>
* cuda: enable topk-moe fusion for 288 experts
The topk-moe fusion only accepted power-of-2 expert counts (or the
special-cased 576), so models with 288 experts (e.g. Step-3.7-Flash)
fell back to the unfused per-layer routing chain: softmax/sigmoid,
argsort, get_rows, sum_rows, div, clamp, scale. At batch size 1 that
is ~330 extra tiny graph nodes per token.
288 is a multiple of the warp size, so the existing kernel already
handles it; this adds the missing template instantiation and accepts
288 in the eligibility check.
Measured on gfx1151 with Step-3.7-Flash IQ4_XS (llama-bench,
-b 4096 -ub 4096 -fa 1 -dio 1 -ctk q8_0 -ctv q8_0; machine idle,
before/after paired so pp4096 stays matched as a load control):
test | before | after
----------------+----------------+----------------
pp4096 | 460.99 ± 0.45 | 462.47 ± 0.34 (unchanged)
tg128 | 19.10 ± 0.04 | 19.56 ± 0.03 (+2.4%)
tg128 @ d30000 | 12.68 ± 0.04 | 12.69 ± 0.03 (unchanged)
Prompt processing is unaffected (the fusion only touches decode
routing). The decode gain is ~+2.4% at shallow context and fades with
depth: by 30k tokens each step is attention-bound over the KV cache,
so removing the fixed routing overhead is no longer visible.
Assisted-By: Claude Fable 5 <noreply@anthropic.com>
* Update tests/test-backend-ops.cpp
Co-authored-by: Oliver Simons <osimons@nvidia.com>
* Add comment for case 288 in topk-moe.cu
---------
Co-authored-by: Oliver Simons <osimons@nvidia.com>
* chat: fix whitespace problems once and for all
* Purge trailing spaces from grammar generation
* Revert "Purge trailing spaces from grammar generation"
This reverts commit b0827ecb7d.
* move conversion code to a dedicated conversion directory and split the files akin to the src/models architecture
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* chat/autoparser: the fixes
* Move optspace() to chat-peg-parser, comment out server tests invalidated due to content now allowed with forced tool calls.
* Trim whitespace on apply instead
* chat: fix parallel_tool_calls default setting based on model capabilities, add tests for parallel tool calls and structured outputs
* Fix ty errors.
* Fix flake8 err
* seems to work
* fix case with new line
Co-authored-by: sayap <sokann@gmail.com>
* gemma 4: fix pre tok regex
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: sayap <sokann@gmail.com>
* Relax prefill parser to allow space.
* Move changes from prefix() to parser generation
* Only allow spaces if we're not having a pure content parser next
* Allow reshuffled arguments in tagged argument parser format tool calls.
* Remove shuffle just keep the optional parsers in any order
* Remove unnecessary import
* Unified delta net handling
* Remove old methods.
* Refactor and optimize
* Adapt autoregressive version from @ymcki
* Change to decay mask approach
* Fix bad permute
* Qwen 3.5 support
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Further fixes
* Use inheritance, remove unneeded conts
* Not like this!
* Remove ggml.h explicit import
* Remove transformers, fix the views
* ACTUALLY fix views, make super calls explicit in conversion.
* Fix conversion again
* Remove extra ggml.h imports
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* It's Qwen3 Next, the lean mean token generation machine!
* Apply patches from thread
* Remove recurrent version, only keep chunked and autoregressive
* Remove unnecessary conts and asserts
* Remove more extra conts and asserts
* Cleanup masking
* Extended TRI
* Fix whitespace
* chore: update webui build output
* Just use cuBLAS for everything...
* Merge both versions
* Remove incorrect imports causing failures for CI
* Still failing... remove all direct cublas imports and rely on common imports from "common.cuh"
* Defines for hipBlas
* Aaaand MUSA defines...
* I hate this job...
* Stupid typo...
* Update ggml/src/ggml-cuda/solve_tri.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Make graph_max_nodes vary by ubatch size for models where chunking might explode the graph
* Update src/llama-context.h
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add missing const
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add support for CUMSUM and TRI for CUDA.
* Minor optimizations.
* Correct warp_prefix_inclusive_sum in float2 variant to return float2
* Optimize TRI
* Whitespace
* Fix strides.
* Implement double loop
* Whitespace
* Fix HIP compilation bugs
* Optimizations + big case performance tests
* Implement using CUB with fallback to custom kernel
* Remove error message.
* Fixes from code review
* Comment out CPU-unsupported F16/BF16 cases to fix CI
* Fine, you win :P
* Fix last cast, use NO_DEVICE_CODE and GGML_UNUSED_VARS
* Vary warp-size based on physical warp size
* Add GGML_UNUSED_VARS in tri as well
* Use constexpr and call prefix_inclusive with warp_size template param
* Update ggml/src/ggml-cuda/cumsum.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Apply suggestions from code review
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Change to tid % warp_size
* Fix strides; hardcode mask; add ggml_lane_mask_t
* Missing renames, remove unused get_warp_mask(), explicit calls to ggml_cuda_info()
* Too hasty...
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Qwen3 Next - cleaned up version
* Whitespaces and stuff
* Correct minor errors
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Misc. fixes.
* Clean up code, add missing hybrid qualifier
* Did someone transpose the SOLVE_TRI result matrix? Perhaps...
* Whitespace
* Proper tensors for cb calls
* Use llama-graph.h vertical alignment
* BROKEN: chunking
* Set new tensors as inputs.
* Proper chunk logic
* It's the circle of life...
* More shenanigans for n_seq > 1
* Nail in the coffin?
* Fix Windows build
* Eh, one fails on Windows, the other fails on Mac... just use general capture.
* quant : cleanup
* model : cleanup
* qwen3 : cleanup
* cont : cleanup
* cont : cleanup
* ggml : revert change
* qwen3 : cleanup
* cont : cleanup
* Readd cmath
* qwen3 : fix typo
* Update convert_hf_to_gguf.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Usual suspects
* fix my bad suggestion
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Fix DoS / integer overflow
* Remove optional, use INT64_MAX instead as placeholder value (it's technically -1, so it fits :)
* White space
* Actually, since it's unsigned, use UINT64_MAX
* Fix too relaxed check on CUDA "fast copy" (can_be_transposed) condition
* Argh.
* Making CISC happy ;)
* Integrate CONT tests
* Use loopy loop
* Skip new tests for (B)F16 for now.
* First attempt
* No permute during convert (fixes qk tensors), proper norm application.
* RoPE = NeoX
* Coherence!
* Migrate xielu params from tensors to hyperparameters
* Simple CUDA kernel
* Revert stupid LLM refactorings
* Chat template support
* configchecker / flake8 errors
* Reorder unary.cu
* I do conclude that LLMs are, in fact, stupid.
* Fix after merge
* Final newline
* Make xIELU an UNARY_OP
* Final newline
* Correctly account for parameter shift
* Argh.
* Update ggml/src/ggml-cpu/unary-ops.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Refactor: remove unused methods, inline and factorize softplus, add const modifiers
* Revert CUDA changes, implement xIELU as a separate OP
* Pesky newline
* Add float2half / half2float for F16 inputs/outputs
* CUDA variants, attempt 2
* Actually, attempt 3
* Update ggml/src/ggml-cuda/unary.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Missing convert header
* Proper formula and reference for xIELU in the comments.
* Modify unary-ops.cpp to add the functor-based logic besides the template system to retain optimizations
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Add tensor mappings for Apertus to global list instead
* Fix lazy on scalars
* Update ggml/src/ggml-cuda/unary.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Add comment about the constraints on positive/negative alpha
* Change `softplus` to `ggml_softplus`
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* feat: Extra debugging support for model conversion - added BF16 support for llama-callback-eval and support for dumping intermediate steps in run-org-model.py
