metadata

license: creativeml-openrail-m
datasets:
  - amphora/QwQ-LongCoT-130K
language:
  - en
base_model: prithivMLmods/QwQ-LCoT-3B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
  - text-generation-inference
  - long-CoT
  - safetensors
  - 3B
  - Instruct
  - QwQ
  - Qwen2.5
  - llama-cpp
  - gguf-my-repo

Triangle104/QwQ-LCoT-3B-Instruct-Q6_K-GGUF

This model was converted to GGUF format from prithivMLmods/QwQ-LCoT-3B-Instruct using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model.

Model details:

The QwQ-LCoT-3B-Instruct model is a lightweight, instruction-tuned language model designed for complex reasoning and explanation tasks. It is fine-tuned on the Qwen2.5-3B-Instruct base model using the QwQ-LongCoT-130K dataset, focusing on long-chain-of-thought (LCoT) reasoning for enhanced logical comprehension and detailed output generation.

Key Features:

Long Chain-of-Thought Reasoning:
    Specifically designed to generate comprehensive, step-by-step explanations for complex queries.

Lightweight and Efficient:
    With only 3 billion parameters, it is optimized for systems with limited computational resources without compromising reasoning capabilities.

Instruction Optimization:
    Fine-tuned to follow prompts and provide concise, actionable, and structured responses.

Training Details:

Base Model: Qwen2.5-3B-Instruct
Dataset: amphora/QwQ-LongCoT-130K
    Comprising 133,000 annotated samples focusing on logical tasks and structured thinking.

Capabilities:

Text Generation:
    Provides detailed, structured, and logical text outputs tailored to user prompts.

Reasoning Tasks:
    Solves step-by-step problems in math, logic, and science.

Educational Assistance:
    Generates coherent explanations for academic and research purposes.

Dialogue and Summarization:
    Handles conversational queries and summarizes long documents effectively.

Usage Instructions:

Setup: Download all model files and ensure compatibility with the Hugging Face Transformers library.

Loading the Model:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/QwQ-LCoT-3B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

Generate Long-Chain Reasoning Outputs:

input_text = "Explain the process of photosynthesis step-by-step." inputs = tokenizer(input_text, return_tensors="pt") outputs = model.generate(**inputs, max_length=300, temperature=0.5) print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Customize Output Generation: Modify the generation_config.json file for different scenarios:

temperature: Controls randomness (lower = deterministic, higher = creative).
max_length: Sets response length.
top_p: Adjusts sampling for diversity in outputs.

Use with llama.cpp

Install llama.cpp through brew (works on Mac and Linux)

brew install llama.cpp

Invoke the llama.cpp server or the CLI.

CLI:

llama-cli --hf-repo Triangle104/QwQ-LCoT-3B-Instruct-Q6_K-GGUF --hf-file qwq-lcot-3b-instruct-q6_k.gguf -p "The meaning to life and the universe is"

Server:

llama-server --hf-repo Triangle104/QwQ-LCoT-3B-Instruct-Q6_K-GGUF --hf-file qwq-lcot-3b-instruct-q6_k.gguf -c 2048

Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well.

Step 1: Clone llama.cpp from GitHub.

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

Step 2: Move into the llama.cpp folder and build it with LLAMA_CURL=1 flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).

cd llama.cpp && LLAMA_CURL=1 make

Step 3: Run inference through the main binary.

./llama-cli --hf-repo Triangle104/QwQ-LCoT-3B-Instruct-Q6_K-GGUF --hf-file qwq-lcot-3b-instruct-q6_k.gguf -p "The meaning to life and the universe is"

./llama-server --hf-repo Triangle104/QwQ-LCoT-3B-Instruct-Q6_K-GGUF --hf-file qwq-lcot-3b-instruct-q6_k.gguf -c 2048