drholy/rkllm-gradio-server
1
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

update models
All checks were successful
Build, Push and Redeploy Docker Image / build-and-push (push) Successful in 5m18s
Details

Browse Source
This commit is contained in:
drholy
2026-02-04 18:59:03 +07:00
parent a66f257fed
commit 17cf7a6a55
1 changed files with 3 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
app/model_configs.py
View File

@@ -15,25 +15,9 @@ model_configs = {
"Qwen2.5-3B-Instruct-w8w8": {"filename": "Qwen2.5-3B-Instruct-w8w8.rkllm"} "Qwen2.5-3B-Instruct-w8w8": {"filename": "Qwen2.5-3B-Instruct-w8w8.rkllm"}
} }
}, },
"DeepSeek-R1-Distill-Qwen-1.5B_FP16_RK3588": { "Qwen3-4B-Instruct-2507-F16": {
"base_config": { "base_config": {
"st_model_id": "Qwen/Qwen2.5-14B-Instruct", "st_model_id": "unsloth/Qwen3-4B-Instruct-2507-GGUF",
"max_context_len": 128000,
"max_new_tokens": 8192,
"top_k": 5,
"top_p": 0.95,
"temperature": 0.6,
"repeat_penalty": 1.00,
"frequency_penalty": 0.2,
"system_prompt": "You are Deepseek. You are a helpful assistant."
},
"models": {
"DeepSeek-R1-Distill-Qwen-1.5B_FP16_RK3588": {"filename": "DeepSeek-R1-Distill-Qwen-1.5B_FP16_RK3588.rkllm"}
}
},
"Qwen2.5-Coder-3B-Instruct-w8w8": {
"base_config": {
"st_model_id": "Qwen/Qwen2.5-14B-Instruct",
"max_context_len": 128000, "max_context_len": 128000,
"max_new_tokens": 8192, "max_new_tokens": 8192,
"top_k": 5, "top_k": 5,
@@ -44,7 +28,7 @@ model_configs = {
"system_prompt": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant" "system_prompt": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant"
}, },
"models": { "models": {
"Qwen2.5-Coder-3B-Instruct-w8w8": {"filename": "Qwen2.5-Coder-3B-Instruct-w8w8.rkllm"} "Qwen3-4B-Instruct-2507-F16": {"filename": "Qwen3-4B-Instruct-2507-F16.rkllm"}
} }
} }
# "LLaMA-Mesh": { # "LLaMA-Mesh": {