m5/StackFlow
0
0
Fork 0
mirror of https://github.com/m5stack/StackFlow.git synced 2026-05-20 11:32:11 -07:00
Code Issues Packages Projects Releases Wiki Activity

[update] update llm-vlm version & model config

Browse Source
This commit is contained in:
LittleMouse
2025-11-21 09:47:03 +08:00
parent 0d3e36fb49
commit 324f04de19
8 changed files with 26 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/llm_framework/main_cosy_voice/src/runner/Tokenizer/Tokenizer.cpp
+1 -1
View File
@@ -389,7 +389,7 @@ public:
bool Init(std::string model_path = "http://localhost:8080") override
{
base_url = model_path;
if (!test_connect_http(base_url, 10)) {
if (!test_connect_http(base_url, 20)) {
ALOGE("connect %s failed", base_url.c_str());
return false;
} else {
projects/llm_framework/main_llm/src/runner/Tokenizer/Tokenizer.cpp
+10 -4
View File
@@ -349,11 +349,17 @@ public:
bool Init(std::string model_path = "http://localhost:8080", bool b_bos = true, bool b_eos = false) override
{
base_url = model_path;
if (!test_connect_http(base_url, 20)) {
ALOGE("connect %s failed", base_url.c_str());
return false;
} else {
ALOGI("connect %s ok", base_url.c_str());
}
try {
cli = std::make_shared<httplib::Client>(base_url);
cli->set_connection_timeout(1);
cli->set_read_timeout(1);
cli->set_write_timeout(1);
cli->set_connection_timeout(10);
cli->set_read_timeout(10);
cli->set_write_timeout(10);
{
auto ret = cli->Get("/bos_id");
auto rep = ret.value();
@@ -389,7 +395,7 @@ public:
bool Init(std::string model_path = "http://localhost:8080") override
{
base_url = model_path;
if (!test_connect_http(base_url, 10)) {
if (!test_connect_http(base_url, 20)) {
ALOGE("connect %s failed", base_url.c_str());
return false;
} else {
projects/llm_framework/main_vlm/SConstruct
+5 -2
View File
@@ -17,7 +17,10 @@ LDFLAGS = []
LINK_SEARCH_PATH = []
STATIC_FILES = []
python_venv = check_wget_down("https://m5stack.oss-cn-shenzhen.aliyuncs.com/resource/linux/llm/m5stack_llm-vlm-python-venv_v1.6.tar.gz", 'm5stack_llm-vlm-python-venv_v1.6.tar.gz')
if 'CONFIG_AX_620E_MSP_ENABLED' in os.environ:
python_venv = check_wget_down("https://m5stack.oss-cn-shenzhen.aliyuncs.com/resource/linux/llm/m5stack_llm-vlm-python-venv_v1.6.tar.gz", 'm5stack_llm-vlm-python-venv_v1.6.tar.gz')
else:
python_venv = check_wget_down("https://m5stack.oss-cn-shenzhen.aliyuncs.com/resource/linux/llm/m5stack_llm-vlm-python-venv_v1.7.tar.gz", 'm5stack_llm-vlm-python-venv_v1.7.tar.gz')
DEFINITIONS += ['-O2']
DEFINITIONS += ['-std=c++17']
@@ -73,7 +76,7 @@ ignore['ignore'] = list(set(ignore['ignore']))
with open('../dist/fileignore', 'w') as f:
json.dump(ignore, f, indent=4)
env['COMPONENTS'].append({'target':'llm_vlm-1.9',
env['COMPONENTS'].append({'target':'llm_vlm-1.10',
'SRCS':SRCS,
'INCLUDE':INCLUDE,
'PRIVATE_INCLUDE':PRIVATE_INCLUDE,
projects/llm_framework/main_vlm/models/mode_qwen3-vl-2B-Int4-ax650.json
+1 -1
View File
@@ -46,7 +46,7 @@
"video_token_id": 151656,
"vision_start_token_id": 151652,
"precompute_len": 0,
"cmm_size": 1919044,
"cmm_size": 3582336,
"ext_scripts": [
"tokenizer_qwen3-vl-2B-Int4-ax650.py"
]
projects/llm_framework/main_vlm/src/main.cpp
+2
View File
@@ -480,6 +480,7 @@ public:
std::vector<std::vector<unsigned short>> all_embeds;
if (auto ret = lLaMa_ctx_->Encode(mats, all_embeds); ret != 0) {
ALOGE("lLaMaCtx.Encode failed");
if (out_callback_) out_callback_("Encode failed", true);
return;
}
mats.clear();
@@ -487,6 +488,7 @@ public:
lLaMa_ctx_->Encode(all_embeds, prompt_data_, prompt_complete(msg), tokens_ids, tokens_diff);
ret != 0) {
ALOGE("lLaMaCtx.Encode failed");
if (out_callback_) out_callback_("Encode failed", true);
return;
}
if (auto ret = lLaMa_ctx_->SetKVCache(k_caches, v_caches, precompute_len, tokens_diff.size());
projects/llm_framework/main_vlm/src/runner/LLM.hpp
+3 -1
View File
@@ -3,6 +3,8 @@
#include <algorithm>
#include <cmath>
#include <numeric>
#include <atomic>
#include "bfloat16.hpp"
#include "image_processor.hpp"
#include "mrope.hpp"
@@ -49,7 +51,7 @@ struct LLMAttrType {
TokenizerType tokenizer_type = TKT_LLaMa;
std::string filename_tokenizer_model = "tokenizer.model";
std::string url_tokenizer_model;
bool b_bos = true;
bool b_bos = false;
bool b_eos = false;
std::string filename_tokens_embed = "tinyllama.model.embed_tokens.weight.bfloat16.bin";
int tokens_embed_num = 32000;
projects/llm_framework/main_vlm/src/runner/Tokenizer/Tokenizer.cpp
+3 -3
View File
@@ -25,7 +25,7 @@ public:
bool Init(std::string model_path) override
{
base_url = model_path;
if (!test_connect_http(base_url, 10)) {
if (!test_connect_http(base_url, 20)) {
ALOGE("connect %s failed", base_url.c_str());
return false;
} else {
@@ -153,7 +153,7 @@ public:
bool Init(std::string model_path = "http://localhost:8080", bool b_bos = true, bool b_eos = false) override
{
base_url = model_path;
if (!test_connect_http(base_url, 10)) {
if (!test_connect_http(base_url, 30)) {
ALOGE("connect %s failed", base_url.c_str());
return false;
} else {
@@ -200,7 +200,7 @@ public:
bool Init_new(std::string model_path, bool b_bos, bool b_eos) override
{
base_url = model_path;
if (!test_connect_http(base_url, 10)) {
if (!test_connect_http(base_url, 30)) {
ALOGE("connect %s failed", base_url.c_str());
return false;
} else {
projects/llm_framework/tools/llm_pack.py
+1 -1
View File
@@ -381,7 +381,7 @@ if __name__ == "__main__":
'llm-tts':[create_bin_deb,'llm-tts', '1.6', src_folder, revision],
'llm-melotts':[create_bin_deb,'llm-melotts', '1.9', src_folder, revision],
'llm-camera':[create_bin_deb,'llm-camera', '1.9', src_folder, revision, 'lib-llm'],
'llm-vlm':[create_bin_deb,'llm-vlm', '1.9', src_folder, revision],
'llm-vlm':[create_bin_deb,'llm-vlm', '1.10', src_folder, revision],
'llm-yolo':[create_bin_deb,'llm-yolo', '1.9', src_folder, revision],
'llm-skel':[create_bin_deb,'llm-skel', version, src_folder, revision],
'llm-depth-anything':[create_bin_deb,'llm-depth-anything', '1.7', src_folder, revision],