CreitinGameplays's picture
Update README.md
fda90a1 verified
|
raw
history blame
1.34 kB
metadata
license: mit
datasets:
  - CreitinGameplays/Magpie-Reasoning-V2-250K-CoT-Deepseek-R1-Llama-70Bmistral
language:
  - en
base_model:
  - mistralai/Mistral-Nemo-Instruct-2407
pipeline_tag: text-generation
library_name: transformers

Mistral Nemo 12B R1

mistralthink

Took 96 hours to finetune on 2x Nvidia RTX A6000 with the following settings:

  • Batch size: 3
  • Gradient accumulation steps: 1
  • Epochs: 1
  • Learning rate: 1e-4
  • Warmup ratio: 0.1

Run the model:

import torch
from transformers import pipeline

model_id = "CreitinGameplays/Mistral-Nemo-12B-R1-v0.1"

pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "How many r's are in strawberry?"}
]

outputs = pipe(
    messages,
    temperature=0.4,
    repetition_penalty=1.1,
    max_new_tokens=2048
)

print(outputs[0]["generated_text"][-1])

System prompt:

You are an AI focused on providing systematic, well-reasoned responses. Response Structure: - Format: <think>{reasoning}</think>{answer} - Process: Think first, then answer. Always use your reasoning capabilities.