vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-08 08:29:11 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Added parameter display for the distilled model of deepseek-qwen

Browse source
This commit is contained in:
leeetao 2025-02-24 13:24:56 +00:00
parent 7bf1b743fb
commit 42da179d66
2 changed files with 14 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
common/common.cpp
View file

@ -873,13 +873,17 @@ static bool assign_layers_to_device(
// model-specific constants
const int n_embd_k_gqa = llama_model_n_embd_k_gqa(model);
const int n_embd_v_gqa = llama_model_n_embd_v_gqa(model);
if (n_embd_k_gqa <= 0 || n_embd_v_gqa <= 0) {
LOG_ERR("Invalid model parameters,n_embd_k_gqa and n_embd_v_gqa cannot be less than 0");
return false;
}
const int n_kv = cparams.n_ctx;
const int64_t b = dev_info_set[0].model_bytes.nb_layer;
const int64_t bo = dev_info_set[0].model_bytes.nb_output;
const int64_t b_prime = b + 2 * (n_embd_k_gqa + n_embd_v_gqa) * n_kv;
#if defined(USE_HIGHS)
#if defined(USE_HIGHS)
const device_info &master = dev_info_set[0];
const int n_vocab = llama_n_vocab(model);
const int64_t bi = dev_info_set[0].model_bytes.nb_input;

10
src/llama.cpp
View file

@ -2318,6 +2318,7 @@ enum e_model {
MODEL_1B,
MODEL_1_3B,
MODEL_1_4B,
MODEL_1_5B,
MODEL_1_6B,
MODEL_2B,
MODEL_2_8B,
@ -2336,6 +2337,7 @@ enum e_model {
MODEL_16B,
MODEL_20B,
MODEL_30B,
MODEL_32B,
MODEL_34B,
MODEL_35B,
MODEL_40B,
@ -5675,6 +5677,7 @@ static const char * llama_model_type_name(e_model type) {
case MODEL_1B: return "1B";
case MODEL_1_3B: return "1.3B";
case MODEL_1_4B: return "1.4B";
case MODEL_1_5B: return "1.5B";
case MODEL_1_6B: return "1.6B";
case MODEL_2B: return "2B";
case MODEL_2_8B: return "2.8B";
@ -5693,6 +5696,7 @@ static const char * llama_model_type_name(e_model type) {
case MODEL_16B: return "16B";
case MODEL_20B: return "20B";
case MODEL_30B: return "30B";
case MODEL_32B: return "32B";
case MODEL_34B: return "34B";
case MODEL_35B: return "35B";
case MODEL_40B: return "40B";
@ -6046,8 +6050,12 @@ static void llm_load_hparams(
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) {
case 24: model.type = hparams.n_embd == 1024 ? e_model::MODEL_0_5B : e_model::MODEL_1B; break;
case 28: model.type = hparams.n_embd == 1536 ? e_model::MODEL_1_5B : e_model::MODEL_7B; break;
case 32: model.type = e_model::MODEL_7B; break;
case 36: model.type = e_model::MODEL_3B; break;
case 40: model.type = hparams.n_head() == 20 ? e_model::MODEL_4B : e_model::MODEL_13B; break;
case 48: model.type = e_model::MODEL_14B; break;
case 64: model.type = e_model::MODEL_32B; break;
case 80: model.type = e_model::MODEL_70B; break;
default: model.type = e_model::MODEL_UNKNOWN;
}
@ -16946,7 +16954,7 @@ static std::vector<struct ggml_cgraph *> llama_build_graph(
} break;
case LLM_ARCH_QWEN2:
{
// result.push_back(llm.build_qwen2()); // TODO:Rewrite the build graph function for distributed inference
// result.push_back(llm.build_qwen2());
result = llm.build_qwen2();
} break;
case LLM_ARCH_QWEN2MOE: