deepfake fake

MT-Bench, VMLU

Convert huggingface model to gguf and build docker image to run with CPU

1. download huggingface model you need, mine is: https://huggingface.co/Viet-Mistral/Vistral-7B-Chat

            git clone https://oauth:..token hf here...@huggingface.co/meta-llama/Meta-Llama-3-8B
   
            git clone https://oauth:..token hf here...@huggingface.co/Viet-Mistral/Vistral-7B-Chat
   
            to folder : "/work/llm/Ner_Llm_Gpt/mistralvn/Vistral-7B-Chat"
   
            git lfs fetch --all
   

2. clone llamacpp: git clone https://github.com/ggerganov/llama.cpp.git

Open terminal , go to inside foler llama.cpp

            cd llama.cpp

            pip install -r requirements.txt

            # build for run in cpu

            mkdir build
            cd build
            cmake ..
            cmake --build . --config Release

3. convert model to gguf

convert.py in folder llama.cpp cloned

            python convert.py "/work/llm/Ner_Llm_Gpt/mistralvn/Vistral-7B-Chat" --outfile Vistral-7B-Chat.gguf --outtype q8_0

4. build docker image and run

            copy build/bin to mistravn/bin (in step 2)
   
            copy Vistral-7B-Chat.gguf to mistravn/Vistral-7B-Chat.gguf (in step 3)
   
            docker build -f dockerfile.llamaccp -t llama-vistral7b .
   
            docker run -d --restart always -p 22222:8880 --name llama-vistral7b_8880 llama-vistral7b
   

5. run bash

            /work/llama.cpp/build/bin/server -m '/work/llama.cpp/Vistral-7B-Chat.gguf' -c 2048 --host 0.0.0.0 --port 8880
   

6. dockerfile.llamacpp

            FROM mcr.microsoft.com/dotnet/aspnet:8.0 AS base
            USER root
            WORKDIR /app 
            RUN apt-get update &&  apt-get install -y build-essential git libc6
            EXPOSE 8880
            COPY /bin/ /app/bin/
            COPY /Vistral-7B-Chat.gguf /app/Vistral-7B-Chat.gguf
            ENV LC_ALL=C.utf8
            CMD [ "/bin/sh", "-c", "./bin/server -m '/app/Vistral-7B-Chat.gguf' -c 2048 --host 0.0.0.0 --port 8880"]
   

7. api restfull call

Keep history chat by newline newline \n\nUser: ...prompt...\nLlama: ...response...\n\n...

            POST: http://localhost:22222/completion

request:

            {"stream":false,"n_predict":400,"temperature":0.7,"stop":["</s>","Llama:","User:"],"repeat_last_n":256,"repeat_penalty":1.18,"penalize_nl":false,"top_k":40,"top_p":0.95,"min_p":0.05,"tfs_z":1,"typical_p":1,"presence_penalty":0,"frequency_penalty":0,"mirostat":0,"mirostat_tau":5,"mirostat_eta":0.1,"grammar":"","n_probs":0,"min_keep":0,"image_data":[],"cache_prompt":true,"api_key":"","prompt":"This is a conversation between User and Llama, a friendly chatbot. Llama is helpful, kind, honest, good at writing, and never fails to answer any requests immediately and with precision.\n\nUser: hi\nLlama: Hello! How may I help you?\n\n\nUser: giới thiệu về Việt Nam\nLlama:"}

response:

            {"content":" Việt Nam là một quốc gia xinh đẹp ở Đông Nam Á với lịch sử và văn hóa phong phú, nổi tiếng với những cảnh quan thiên nhiên tuyệt vời như Vịnh Hạ Long hay động Phong Nha. Nó cũng có nền ẩm thực đa dạng phản ánh di sản lâu đời của nó ","id_slot":0,"stop":true,"model":"/app/Vistral-7B-Chat.gguf","tokens_predicted":58,"tokens_evaluated":78,"generation_settings":{"n_ctx":2048,"n_predict":-1,"model":"/app/Vistral-7B-Chat.gguf","seed":4294967295,"temperature":0.699999988079071,"dynatemp_range":0.0,"dynatemp_exponent":1.0,"top_k":40,"top_p":0.949999988079071,"min_p":0.05000000074505806,"tfs_z":1.0,"typical_p":1.0,"repeat_last_n":256,"repeat_penalty":1.1799999475479126,"presence_penalty":0.0,"frequency_penalty":0.0,"penalty_prompt_tokens":[],"use_penalty_prompt_tokens":false,"mirostat":0,"mirostat_tau":5.0,"mirostat_eta":0.10000000149011612,"penalize_nl":false,"stop":["</s>","Llama:","User:"],"n_keep":0,"n_discard":0,"ignore_eos":false,"stream":false,"logit_bias":[],"n_probs":0,"min_keep":0,"grammar":"","samplers":["top_k","tfs_z","typical_p","top_p","min_p","temperature"]},"prompt":"This is a conversation between User and Llama, a friendly chatbot. Llama is helpful, kind, honest, good at writing, and never fails to answer any requests immediately and with precision.\n\nUser: hi\nLlama: Hello! How may I help you?\n\n\nUser: giới thiệu về Việt Nam\nLlama:","truncated":false,"stopped_eos":true,"stopped_word":false,"stopped_limit":false,"stopping_word":"","tokens_cached":135,"timings":{"prompt_n":1,"prompt_ms":196.862,"prompt_per_token_ms":196.862,"prompt_per_second":5.079700500858469,"predicted_n":58,"predicted_ms":11245.507,"predicted_per_token_ms":193.88805172413794,"predicted_per_second":5.157615392529657}}

8. docker nvidia gpu build

NVIDIA : https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_local

            mkdir build
            cd build
            export CUDACXX=/usr/local/cuda-12/bin/nvcc
            cmake .. -DLLAMA_CUDA=ON -DCMAKE_CUDA_COMPILER=/usr/local/cuda-12/bin/nvcc -DCUDAToolkit_ROOT=/usr/local/cuda-12
            cmake --build . --config Release

             python "/work/llm/llama.cpp/convert.py" "/work/llm/Ner_Llm_Gpt/mistralvn/Vistral-7B-Chat" --outfile "/work/llm/Ner_Llm_Gpt/mistralvn/Vistral-7B-Chat.gpu.gguf" --outtype q8_0

dockerfile

            curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
            curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
            sudo apt-get update
            sudo apt-get install -y nvidia-docker2

            sudo systemctl daemon-reload
            sudo systemctl restart docker

            docker pull nvidia/cuda:12.4.1-devel-ubuntu22.04 

dockerfile.gpu.llamaccp

            FROM nvidia/cuda:12.4.1-devel-ubuntu22.04 
            USER root
            WORKDIR /app 
            EXPOSE 8880
            COPY /bin.gpu/ /app/bin/

            COPY /favicon.ico /app/favicon.ico
            COPY /favicon.png /app/favicon.png

            COPY /Vistral-7B-Chat.gpu.gguf /app/Vistral-7B-Chat.gguf
            ENV LLAMA_CUDA=1
            ENV LLAMA_CURL=1
            ENV CUDA_DOCKER_ARCH=all
            ENV LC_ALL=C.utf8
            CMD [ "/bin/sh", "-c", "./bin/server -m '/app/Vistral-7B-Chat.gguf' -c 2048 --host 0.0.0.0 --port 8880"]

9. docker amd rocm gpu build

finetune vistral

ref: https://blog.ngxson.com/fine-tune-model-viet-truyen-phong-cach-nam-cao/

Ner_Llm_Gpt

Support Vietnamese

https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0

llamma.cpp dockerfile

more usage: https://github.com/badpaybad/llama.cpp.docker

                # if manual download just uncomment this , and comment curl
                COPY llava-v1.5-7b-q4-server.llamafile /app/llava-v1.5-7b-q4-server.llamafile

                # RUN curl -LO https://huggingface.co/jartine/llava-v1.5-7B-GGUF/resolve/main/llava-v1.5-7b-q4-server.llamafile
                RUN chmod 755 /app/llava-v1.5-7b-q4-server.llamafile


                #sudo docker build --no-cache -f dockerfile -t docker.io/dunp/llamacpp .
                #docker run  -d --restart always -p 8080:8080 --name llamacpp_8880 docker.io/dunp/llamacpp


                #https://github.com/karpathy/llama2.c 

hugginface git access token

                https://huggingface.co/settings/profile

                https://huggingface.co/settings/tokens 

vinallamaGPT

                git clone https://huggingface.co/vilm/vinallama-2.7b     

                git clone https://oauth:hf_...yourtoken....@huggingface.co/vilm/vinallama-7b  

phoGPT

                git clone https://huggingface.co/vinai/PhoGPT-7B5-Instruct
                git clone https://oauth:hf_...yourtoken....@huggingface.co/vinai/PhoGPT-7B5-Instruct/