EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

Support diffusion models: Add Dream 7B (#14644)

Browse Source 
* Support diffusion models: Add Dream 7B

* Move diffusion to examples

* Move stuff to examples. Add patch to not use kv-cache

* Address review comments

* Make sampling fast

* llama: remove diffusion functions

* Add basic timings + cleanup

* More cleanup

* Review comments: better formating, use LOG instead std::cerr, re-use batch, use ubatch instead of max_length

* fixup!

* Review: move everything to diffusion-cli for now
This commit is contained in:
Aman Gupta
2025-07-16 20:03:51 +08:00
committed by GitHub
parent 64978340b0
commit ab14019821
13 changed files with 804 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/llama-arch.cpp
View File

@@ -85,6 +85,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_HUNYUAN_MOE, "hunyuan-moe" },
{ LLM_ARCH_SMOLLM3, "smollm3" },
{ LLM_ARCH_LFM2, "lfm2" },
{ LLM_ARCH_DREAM, "dream" },
{ LLM_ARCH_UNKNOWN, "(unknown)" },
};
@@ -1891,6 +1892,23 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
},
},
{
LLM_ARCH_DREAM,
{
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
{ LLM_TENSOR_OUTPUT_NORM, "output_norm" },
{ LLM_TENSOR_OUTPUT, "output" },
{ LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
{ LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" },
{ LLM_TENSOR_ATTN_K, "blk.%d.attn_k" },
{ LLM_TENSOR_ATTN_V, "blk.%d.attn_v" },
{ LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" },
{ LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" },
{ LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" },
{ LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" },
{ LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
},
},
};
static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
@@ -2133,3 +2151,12 @@ bool llm_arch_is_hybrid(const llm_arch & arch) {
return false;
}
}
bool llm_arch_is_diffusion(const llm_arch & arch) {
switch (arch) {
case LLM_ARCH_DREAM:
return true;
default:
return false;
}
}