Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

DeepseekV4: fix rollback with multi-seq (#26756)

Browse Source 
* DeepseekV4: fix rollback with multi-seq

* fix model loading

* make pending rollback single use

* only clear cache for seq_id for full load

* add assert for compress ratio

* make graph topology static

* pass true instead of flags in clear_compressed

* cont : clean-up + TODOs

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Aman Gupta
2026-08-23 13:57:49 +03:00
committed by GitHub
co-authored by Georgi Gerganov
parent d3371929bb
commit b0539c43ed
10 changed files with 345 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model-saver.cpp
+11
View File
@@ -293,6 +293,14 @@ void llama_model_saver::add_kv_from_model() {
add_kv(LLM_KV_ATTENTION_INDEXER_LOCAL_BLOCKS, hparams.indexer_local_blocks);
add_kv(LLM_KV_ATTENTION_INDEXER_TYPES, hparams.is_indexer_full_impl, true);
add_kv(LLM_KV_ATTENTION_RECURRENT_LAYERS, hparams.is_recr_impl, true);
add_kv(LLM_KV_ATTENTION_OUTPUT_GROUP_COUNT, hparams.dsv4_o_group_count);
add_kv(LLM_KV_ATTENTION_OUTPUT_LORA_RANK, hparams.dsv4_o_lora_rank);
add_kv(LLM_KV_ATTENTION_COMPRESS_ROPE_FREQ_BASE, hparams.dsv4_compress_rope_base);
add_kv(LLM_KV_ATTENTION_COMPRESS_RATIOS, hparams.dsv4_compress_ratios, true);
add_kv(LLM_KV_HYPER_CONNECTION_COUNT, hparams.dsv4_hc_mult);
add_kv(LLM_KV_HYPER_CONNECTION_SINKHORN_ITERATIONS, hparams.dsv4_hc_sinkhorn_iters);
add_kv(LLM_KV_HYPER_CONNECTION_EPSILON, hparams.dsv4_hc_eps);
add_kv(LLM_KV_HASH_LAYER_COUNT, hparams.dsv4_hash_layer_count);
const float rope_scaling_factor = hparams.rope_freq_scale_train == 1.0f ? 0.0f : 1.0f/hparams.rope_freq_scale_train;
@@ -422,6 +430,9 @@ void llama_model_saver::add_tensors_from_model() {
add_tensor(model->cls_out);
add_tensor(model->cls_out_b);
add_tensor(model->cls_norm);
add_tensor(model->hc_head_fn);
add_tensor(model->hc_head_base);
add_tensor(model->hc_head_scale);
for (const struct llama_layer & layer : model->layers) {
for (size_t i = 0; i < sizeof(layer)/sizeof(struct ggml_tensor *); ++i) {