DeepSeek V4 (#24162)
* convert: add dsv4 conversion * add basic setup * add llm_graph_input_dsv4 * add save-load state * add sinkhorn eps - correction by @fairydreaming * add rope fix * cleanup dead code * fix bugs * support pro model: added by @fairydreaming * remove redundant V cache * Chat template * remove debugging leftovers * Add mechanism for inlining templates based on architecture * s/deepseek-v4-flash/deepseek4/g * s/deepseek-v4-flash/deepseek4/g continued * enable graph reuse * enable FA * fix test llama archs * rename * compatibility with antirez ds4 GGUFs * simplified set_gguf_parameters() by calling super class method, replaced moe.score_func with expert_gating_func. * reserve worst-case kv-cache * revert max split inputs * address review comments * add padding to enable FA * pad only the final value of plan.n_kv to 256 * remove built-in cpp chat template * cont: remove cpp built-in template * rm outdated test * replace ggml_view_3d() with ggml_reshape_3d() Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * only support n_seq=1 for now * remove unused var * cont: remove unused var * use scale bias * use correct ptr for can_reuse * remove gen-chat-inline-templates.py * simplify graph reuse * cont: cleanup * remove unused inputs * enable partial checkpointing * add correct shape for kq_mask + set llama_model_n_swa to 0 for dsv4 * precompute source_idx + add comment about dummy write * support multi-seq * remove restored_trim_pos * use split_equal when possible * fix indent * address review comments * use LLM_KV * fix ci --------- Co-authored-by: Piotr Wilkin <piotr.wilkin@syndatis.com> Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com> Co-authored-by: Xuan Son Nguyen <son@huggingface.co> Co-authored-by: fairydreaming <166155368+fairydreaming@users.noreply.github.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Aman Gupta
co-authored by
Georgi Gerganov
Piotr Wilkin
Stanisław Szymczyk
Xuan Son Nguyen
fairydreaming
Georgi Gerganov
Piotr Wilkin
Stanisław Szymczyk
Xuan Son Nguyen
fairydreaming
parent
6cb18b2f2e
commit
8c146a8366
File diff suppressed because it is too large
Load Diff
@@ -1085,6 +1085,121 @@ struct llama_model_deepseek32 : public llama_model_base {
|
||||
};
|
||||
|
||||
|
||||
struct llama_model_deepseek4 : public llama_model_base {
|
||||
llama_model_deepseek4(const struct llama_model_params & params) : llama_model_base(params) {}
|
||||
void load_arch_hparams(llama_model_loader & ml) override;
|
||||
void load_arch_tensors(llama_model_loader & ml) override;
|
||||
|
||||
struct graph : public llm_graph_context {
|
||||
graph(const llama_model & model, const llm_graph_params & params);
|
||||
|
||||
ggml_tensor * build_hc_pre(
|
||||
ggml_tensor * x,
|
||||
ggml_tensor * hc_fn,
|
||||
ggml_tensor * hc_scale,
|
||||
ggml_tensor * hc_base,
|
||||
ggml_tensor ** post,
|
||||
ggml_tensor ** comb,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hc_post(
|
||||
ggml_tensor * x,
|
||||
ggml_tensor * residual,
|
||||
ggml_tensor * post,
|
||||
ggml_tensor * comb,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hc_head(
|
||||
ggml_tensor * x,
|
||||
ggml_tensor * hc_fn,
|
||||
ggml_tensor * hc_scale,
|
||||
ggml_tensor * hc_base) const;
|
||||
|
||||
ggml_tensor * build_attention(
|
||||
const llama_model & model,
|
||||
llm_graph_input_dsv4 * inp_dsv4,
|
||||
ggml_tensor * cur,
|
||||
ggml_tensor * inp_pos,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hca_compressed_kv_from_state(
|
||||
ggml_tensor * kv_state,
|
||||
ggml_tensor * score_state,
|
||||
ggml_tensor * state_read_idxs,
|
||||
ggml_tensor * comp_pos,
|
||||
ggml_tensor * norm,
|
||||
int64_t n_embd_head,
|
||||
const char * name,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_overlap_compressed_kv_from_state(
|
||||
ggml_tensor * kv_state,
|
||||
ggml_tensor * score_state,
|
||||
ggml_tensor * state_read_idxs,
|
||||
ggml_tensor * comp_pos,
|
||||
ggml_tensor * norm,
|
||||
int64_t ratio,
|
||||
int64_t n_embd_head,
|
||||
const char * name,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_lid_top_k(
|
||||
const llama_model & model,
|
||||
llm_graph_input_dsv4 * inp_dsv4,
|
||||
ggml_tensor * qr,
|
||||
ggml_tensor * cur,
|
||||
ggml_tensor * inp_pos,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_top_k_mask(
|
||||
ggml_tensor * kq_mask,
|
||||
ggml_tensor * top_k,
|
||||
const char * name,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_csa_lid_attention(
|
||||
const llama_model & model,
|
||||
llm_graph_input_dsv4 * inp_dsv4,
|
||||
llm_graph_input_dsv4_raw * inp_attn,
|
||||
ggml_tensor * q,
|
||||
ggml_tensor * kv,
|
||||
ggml_tensor * qr,
|
||||
ggml_tensor * cur,
|
||||
ggml_tensor * inp_pos,
|
||||
ggml_tensor * sinks,
|
||||
float kq_scale,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hca_attention(
|
||||
llm_graph_input_dsv4 * inp_dsv4,
|
||||
llm_graph_input_dsv4_raw * inp_attn,
|
||||
ggml_tensor * q,
|
||||
ggml_tensor * kv,
|
||||
ggml_tensor * sinks,
|
||||
float kq_scale,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_raw_attention(
|
||||
llm_graph_input_dsv4_raw * inp_attn,
|
||||
ggml_tensor * q,
|
||||
ggml_tensor * kv,
|
||||
ggml_tensor * sinks,
|
||||
float kq_scale,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hc_weighted_sum(
|
||||
ggml_tensor * x,
|
||||
ggml_tensor * weights) const;
|
||||
|
||||
ggml_tensor * build_hc_sinkhorn(
|
||||
ggml_tensor * comb,
|
||||
int il) const;
|
||||
};
|
||||
|
||||
std::unique_ptr<llm_graph_context> build_arch_graph(const llm_graph_params & params) const override;
|
||||
};
|
||||
|
||||
|
||||
struct llama_model_deepseek2ocr : public llama_model_base {
|
||||
llama_model_deepseek2ocr(const struct llama_model_params & params) : llama_model_base(params) {}
|
||||
void load_arch_hparams(llama_model_loader & ml) override;
|
||||
|
||||
Reference in New Issue
Block a user