Aman GuptaandGeorgi Gerganov a074215352 deepseekv4.1: follow-up review fixes (#31137) ...

* kv-cache : use ggml_backend_tensor_copy to refill the DSV4 K mirrors

* llama : move the Engram pad id to llama_hparams and the token map to llama_model

The saver no longer needs the deepseek41 model class.

* deepseek41 : drop the F32 src1 request on the HC mixes matmul

* ggml : add softplus TODOs in the backends that still use log(1 + exp(x))

* convert : make the DSV4/DSV4.1 modify_tensors generators, drop the redundant vision filter

* deepseek41 : clean up comments

* deepseek41 : drop the F32 src1 requests on the router and indexer matmuls

* convert : use format_tensor_name directly in DeepseekV4Model

* s/comp_kv_src/comp_il_src/g

* cont : fix TODOs for log1p

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

2026-10-26 15:52:02 +08:00

..

models

deepseekv4.1: follow-up review fixes (#31137)

2026-10-26 15:52:02 +08:00

CMakeLists.txt

llama : add a GPU cache for MoE experts kept in host memory (#29887)

2026-10-07 21:07:45 +03:00

llama-adapter.cpp

lora : support prism.hadamard folded weights (#30910)

2026-10-22 08:12:17 +03:00

llama-adapter.h

llama : re-enable manual LoRA adapter free (#19983)

2026-03-18 12:03:26 +02:00

llama-arch.cpp

qwen: merge tensors on convert (#31028)

2026-10-25 15:23:22 +02:00

llama-arch.h

qwen: merge tensors on convert (#31028)

2026-10-25 15:23:22 +02:00

llama-batch.cpp

spec : add lilicorr support (#30506)

2026-10-15 05:49:25 +02:00

llama-batch.h

model: fix vllm d3 multi seq case (#30408)

2026-10-13 00:23:01 +02:00

llama-chat.cpp

chat : add Granite 4.1 chat template (#23518)

2026-05-28 13:13:33 +02:00

llama-chat.h

chat : add Granite 4.1 chat template (#23518)

2026-05-28 13:13:33 +02:00

llama-context.cpp

model: DSV4.1 flash (#31086)

2026-10-25 09:43:51 +08:00

llama-context.h

llama : skip logits buffer allocation when graph produces no logits (#30255)

2026-10-13 11:24:20 +03:00

llama-cparams.cpp

cparams : rename LLAMA_MAX_PARALLEL_SEQUENCES to LLAMA_MAX_SEQ (#14188)

2025-06-15 10:08:58 +03:00

llama-cparams.h

llama: create new fused op for QSA mask (#30640)

2026-10-22 01:02:22 +08:00

llama-ext.h

model : add H2O-Lightning-4B decision model (#30361)

2026-10-20 18:54:27 +02:00

llama-grammar.cpp

common, grammar: add recursive depth limit, make it a requirement (#31153)

2026-10-25 23:53:08 +01:00

llama-grammar.h

grammar: limit recursion depth to prevent stack overflow (#29495) (#29755)

2026-10-25 12:10:10 +01:00

llama-graph.cpp

model: DSV4.1 flash (#31086)

2026-10-25 09:43:51 +08:00

llama-graph.h

qwen4exp : build the QSA selection without the kq_mask (#30965)

2026-10-23 19:49:08 +02:00

llama-hparams.cpp

llama: add llama_batch_ext (#24669)

2026-09-24 16:25:07 +02:00

llama-hparams.h

deepseekv4.1: follow-up review fixes (#31137)

2026-10-26 15:52:02 +08:00

llama-impl.cpp

llama : optimize vocabulary loading (#30679)

2026-10-21 12:03:27 +03:00

llama-impl.h

llama : optimize vocabulary loading (#30679)

2026-10-21 12:03:27 +03:00

llama-io.cpp

server : avoid checkpoint data host copies (#22558)

2026-05-02 18:03:25 +03:00

llama-io.h

llama : fix K/V and recurrent state cleanup after failed restores (#27530)

2026-09-26 10:23:03 +03:00

llama-kv-cache-dsa-iswa.cpp

kv-cache: honor PARTIAL_ONLY in the plain KV cache (#30514)

2026-10-15 07:39:18 +03:00

llama-kv-cache-dsa-iswa.h

model: add dots3-note (#27060)

2026-08-21 19:52:34 +02:00

llama-kv-cache-dsa.cpp

llama : fix K/V and recurrent state cleanup after failed restores (#27530)

2026-09-26 10:23:03 +03:00

llama-kv-cache-dsa.h

llama : allocate indexer cache only in "full" indexer layers (#26474)

2026-08-03 14:56:30 +02:00

llama-kv-cache-dsv4.cpp

deepseekv4.1: follow-up review fixes (#31137)

2026-10-26 15:52:02 +08:00

llama-kv-cache-dsv4.h

deepseekv4.1: follow-up review fixes (#31137)

2026-10-26 15:52:02 +08:00

llama-kv-cache-iswa.cpp

kv-cache: honor PARTIAL_ONLY in the plain KV cache (#30514)

2026-10-15 07:39:18 +03:00

llama-kv-cache-iswa.h

DeepSeek V4 (#24162)

2026-06-29 16:58:51 +08:00

llama-kv-cache-msa.cpp

model: M3: Move MSA into a new memory implementation (#26338)

2026-08-03 16:30:08 +03:00

llama-kv-cache-msa.h

model: M3: Move MSA into a new memory implementation (#26338)

2026-08-03 16:30:08 +03:00

llama-kv-cache.cpp

model: DSV4.1 flash (#31086)

2026-10-25 09:43:51 +08:00

llama-kv-cache.h

kv-cache: fix restoring mismatched KV cache rotation by saving exact rotation metadata (#28498)

2026-10-05 11:27:30 +03:00

llama-kv-cells.h

add GLM-5.3-Flash (GLM5-Next) support (#27773)

2026-09-30 14:20:32 +08:00

llama-memory-hybrid-idx.cpp

glm5 : optimize inputs (#30932)

2026-10-22 16:59:29 +03:00

llama-memory-hybrid-idx.h

llama: remove the gather path of the glm5-next sparse attention (#30042)

2026-10-06 18:31:13 +02:00

llama-memory-hybrid-iswa.cpp

llama-batch: add n_keep_tail in split_equal for recurrent models (#25278)

2026-07-08 15:55:19 +08:00

llama-memory-hybrid-iswa.h

llama + spec: MTP Support (#22673)

2026-05-16 20:06:23 +08:00

llama-memory-hybrid.cpp

kv-cache: honor PARTIAL_ONLY in the plain KV cache (#30514)

2026-10-15 07:39:18 +03:00

llama-memory-hybrid.h

llama + spec: MTP Support (#22673)

2026-05-16 20:06:23 +08:00

llama-memory-recurrent.cpp

llama : respect no_alloc in all memory caches (#31087)

2026-10-25 11:56:04 +02:00

llama-memory-recurrent.h

Qwen4Exp: add MTP (#29761)

2026-10-01 14:13:27 +03:00

llama-memory.cpp

tests : consolidate context state tests into test-llama-context (#30023)

2026-10-13 16:21:55 +03:00

llama-memory.h

model: fix vllm d3 multi seq case (#30408)

2026-10-13 00:23:01 +02:00

llama-mmap.cpp

llama : split llama_prefetch() into merge and prefetch steps (#30532)

2026-10-15 09:12:45 +02:00

llama-mmap.h

llama-mmap : avoid a second full-size copy of each tensor with direct-io (#29749)

2026-10-01 09:44:31 +02:00

llama-model-loader.cpp

model: DSV4.1 flash (#31086)

2026-10-25 09:43:51 +08:00

llama-model-loader.h

llama : do not map foreign ranges into a backend buffer (#30405)

2026-10-13 17:28:15 +03:00

llama-model-saver.cpp

deepseekv4.1: follow-up review fixes (#31137)

2026-10-26 15:52:02 +08:00

llama-model-saver.h

model: add Qwen3.8-Flash-Next (qwen4exp) (#27742)

2026-08-27 21:32:31 +02:00

llama-model.cpp

model: DSV4.1 flash (#31086)

2026-10-25 09:43:51 +08:00

llama-model.h

deepseekv4.1: follow-up review fixes (#31137)

2026-10-26 15:52:02 +08:00

llama-moe-cache.cpp

moe-cache : replace MoE LRU cache with a 50/50 SLRU cache (#30637)

2026-10-19 13:28:50 +02:00

llama-moe-cache.h

moe-cache : replace MoE LRU cache with a 50/50 SLRU cache (#30637)

2026-10-19 13:28:50 +02:00

llama-quant.cpp

qwen: merge tensors on convert (#31028)

2026-10-25 15:23:22 +02:00

llama-quant.h

llama : refactor src/llama.cpp (#10902)

2025-01-03 10:18:53 +02:00

llama-sampler.cpp

llama: fix exhausting CPU and MEM caused by dry sequence breaker (#31035)

2026-10-25 12:31:09 +01:00

llama-sampler.h

llama : support multi-output backend sampling (#25532)

2026-08-10 16:58:56 +03:00

llama-version.h.in

llama: fix whole source code rebuilt on each new commit (#28278)

2026-09-03 23:53:04 +02:00

llama-vocab.cpp

common, grammar: add recursive depth limit, make it a requirement (#31153)

2026-10-25 23:53:08 +01:00

llama-vocab.h

mtmd : add Soprano TTS support (#28607)

2026-10-24 01:09:50 +02:00

llama.cpp

llama : use GGML_ABORT instead of throw (#29840)

2026-10-02 10:44:44 +02:00

unicode-data.cpp

unicode : add opt-in NFC normalization for Unigram tokenizers (#30390)

2026-10-14 08:33:24 +03:00

unicode-data.h

unicode : add opt-in NFC normalization for Unigram tokenizers (#30390)

2026-10-14 08:33:24 +03:00

unicode.cpp

llama : optimize vocabulary loading (#30679)

2026-10-21 12:03:27 +03:00

unicode.h

llama : optimize vocabulary loading (#30679)

2026-10-21 12:03:27 +03:00