Instructions to use fblgit/TheBeagle-v2beta-32B-MGS with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use fblgit/TheBeagle-v2beta-32B-MGS with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="fblgit/TheBeagle-v2beta-32B-MGS") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("fblgit/TheBeagle-v2beta-32B-MGS") model = AutoModelForCausalLM.from_pretrained("fblgit/TheBeagle-v2beta-32B-MGS", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use fblgit/TheBeagle-v2beta-32B-MGS with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "fblgit/TheBeagle-v2beta-32B-MGS" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "fblgit/TheBeagle-v2beta-32B-MGS", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/fblgit/TheBeagle-v2beta-32B-MGS
- SGLang
How to use fblgit/TheBeagle-v2beta-32B-MGS with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "fblgit/TheBeagle-v2beta-32B-MGS" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "fblgit/TheBeagle-v2beta-32B-MGS", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "fblgit/TheBeagle-v2beta-32B-MGS" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "fblgit/TheBeagle-v2beta-32B-MGS", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use fblgit/TheBeagle-v2beta-32B-MGS with Docker Model Runner:
docker model run hf.co/fblgit/TheBeagle-v2beta-32B-MGS
cool model guys!
i was able to test the Q8 the result is really great!!
i checked this model with some coding task it did really well it would be nice if this models gets tuned for coding task.
update!!
i used this model in web ui with parameter called Contrastive Search which yield higher accuracy on coding task. i will do more test in coming months.
i can give it a shot to try SFT it with some extra coding dataset, keep tuned. and again ,thanks for the feedback.
sure !
i am using this prompt for testing coding but the model not able to identityfi the coding and question when i use this prompt. its more simlar to cot prompt
You should must follow this 6 steps :
Step 1: Understand the Question
Objective: Carefully read and comprehend the core of the question or problem.
Actions:
*Identify the main topic or issue.
*Note any specific details or requirements.
*Determine the type of response expected (e.g., a detailed explanation, a solution, a comparison).
Step 2: Clarify with Sub-Questions
Objective: Ensure a full understanding of the context and requirements.
Actions:
*Ask clarifying questions if needed.
*Seek additional information to fill any gaps in understanding.
*Confirm any assumptions or interpretations with the person asking the question.
Step 3: Research Thoroughly
Objective: Gather accurate and relevant information from reliable sources.
Actions:
*Identify credible sources (e.g., academic journals, reputable websites, expert opinions).
*Collect data and information that directly relates to the question.
*Verify the accuracy and reliability of the information.
Step 4: Break Down the Research
Objective: Organize the gathered information into key points and categories for clarity.
Actions:
*Summarize the main findings.
*Categorize information into relevant sections or themes.
*Highlight key facts, statistics, and supporting evidence.
Step 5: Plan the Answer
Objective: Structure the response logically, ensuring it addresses all aspects of the question.
Actions:
*Outline the main points you will cover.
*Organize the information in a logical sequence.
*Ensure each part of the question is addressed comprehensively.
Step 6: Provide the Answer
Objective: Deliver a well-structured, comprehensive, and clear answer based on the research and planning.
Actions:
*Write or present the answer in a clear and concise manner.
*Use appropriate language and tone for the audience.
*Include supporting evidence and examples where relevant.
*Review the answer to ensure it is complete and accurate.
@MaziyarPanahi please check this model and training method i feel this one is performing really well.
are u using quants or fullweights? the quants has not been updated yet with the new chat template that i uploaded yesterday that boosts IFeval and Math substantially :)
@MaziyarPanahi please check this model and training method i feel this one is performing really well.
very nice! great job! will check it out for sure :)
@fblgit i tested the qunat from https://huggingface.co/fblgit/TheBeagle-v2beta-MGS-GGUF/tree/main
but its looks the templet worngly updated and also its in .bin formate + only able check Q6.
q6 doing really fine by using the Contrastive Search.
if possible please upload Q8 in gguf formate.