download
git clone https://github.com/jingyaogong/minimind.git
huggingface-cli download --local-dir MiniMind2
test
python eval_model.py --load 1 --model_mode 2
cd scripts
streamlit run web_demo.py
使用llama.cpp推理
在llama.cpp中打上如下patch:
diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index bf6bc683..5ea853b6 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -808,7 +808,8 @@ class TextModel(ModelBase):
logger.warning(f"** chkhsh: {chkhsh}")
logger.warning("**************************************************************************************")
logger.warning("\n")
- raise NotImplementedError("BPE pre-tokenizer was not recognized - update get_vocab_base_pre()")
+ res = "smollm"
+ #raise NotImplementedError("BPE pre-tokenizer was not recognized - update get_vocab_base_pre()")
logger.debug(f"tokenizer.ggml.pre: {repr(res)}")
logger.debug(f"chkhsh: {chkhsh}")
python convert_hf_to_gguf.py ../minimind/MiniMind2/
./build/bin/llama-quantize ../minimind/MiniMind2/MiniMind2-109M-F16.gguf ../minimind/MiniMind2/Q4-MiniMind2.gguf Q4_K_M
./build/bin/llama-cli -m ../minimind/MiniMind2/MiniMind2-109M-F16.gguf --chat-template chatml
使用ollama部署
新建minimind.modelfile：
FROM ./MiniMind2-109M-F16.gguf
TEMPLATE """{{ if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}{{ if .Prompt }}<|im_start|>user
{{ .Prompt }}<|im_end|>
{{ end }}<|im_start|>assistant
"""
加载模型并命名为minimind2
ollama create -f minimind.modelfile minimind2
启动推理
ollama run minimind2