metadata

license: apache-2.0
language:
  - en
base_model:
  - Qwen/Qwen2.5-7B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
  - LCoT
  - Qwen
  - v2
datasets:
  - PowerInfer/QWQ-LONGCOT-500K
  - AI-MO/NuminaMath-CoT
  - prithivMLmods/Math-Solve
  - amphora/QwQ-LongCoT-130K
  - prithivMLmods/Deepthink-Reasoning
model-index:
  - name: QwQ-LCoT2-7B-Instruct
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: wis-k/instruction-following-eval
          split: train
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 55.76
            name: averaged accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FQwQ-LCoT2-7B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: SaylorTwift/bbh
          split: test
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 34.37
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FQwQ-LCoT2-7B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: lighteval/MATH-Hard
          split: test
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 22.21
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FQwQ-LCoT2-7B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          split: train
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 6.38
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FQwQ-LCoT2-7B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 15.75
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FQwQ-LCoT2-7B-Instruct
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 37.13
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FQwQ-LCoT2-7B-Instruct
          name: Open LLM Leaderboard

QwQ-LCoT2-7B-Instruct

The QwQ-LCoT2-7B-Instruct is a fine-tuned language model designed for advanced reasoning and instruction-following tasks. It leverages the Qwen2.5-7B base model and has been fine-tuned on the chain of thought reasoning datasets, focusing on chain-of-thought (CoT) reasoning for problems. This model is optimized for tasks requiring logical reasoning, detailed explanations, and multi-step problem-solving, making it ideal for applications such as instruction-following, text generation, and complex reasoning tasks.

Quickstart with Transformers

Here provides a code snippet with apply_chat_template to show you how to load the tokenizer and model and how to generate contents.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/QwQ-LCoT2-7B-Instruct"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "How many r in strawberry."
messages = [
    {"role": "system", "content": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

Intended Use

The QwQ-LCoT2-7B-Instruct model is designed for advanced reasoning and instruction-following tasks, with specific applications including:

Instruction Following: Providing detailed and step-by-step guidance for a wide range of user queries.
Logical Reasoning: Solving problems requiring multi-step thought processes, such as math problems or complex logic-based scenarios.
Text Generation: Crafting coherent, contextually relevant, and well-structured text in response to prompts.
Problem-Solving: Analyzing and addressing tasks that require chain-of-thought (CoT) reasoning, making it ideal for education, tutoring, and technical support.
Knowledge Enhancement: Leveraging reasoning datasets to offer deeper insights and explanations for a wide variety of topics.

Limitations

Data Bias: As the model is fine-tuned on specific datasets, its outputs may reflect inherent biases from the training data.
Context Limitation: Performance may degrade for tasks requiring knowledge or reasoning that significantly exceeds the model's pretraining or fine-tuning context.
Complexity Ceiling: While optimized for multi-step reasoning, exceedingly complex or abstract problems may result in incomplete or incorrect outputs.
Dependency on Prompt Quality: The quality and specificity of the user prompt heavily influence the model's responses.
Non-Factual Outputs: Despite being fine-tuned for reasoning, the model can still generate hallucinated or factually inaccurate content, particularly for niche or unverified topics.
Computational Requirements: Running the model effectively requires significant computational resources, particularly when generating long sequences or handling high-concurrency workloads.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here! Summarized results can be found here!

Metric	Value (%)
Average	28.60
IFEval (0-Shot)	55.76
BBH (3-Shot)	34.37
MATH Lvl 5 (4-Shot)	22.21
GPQA (0-shot)	6.38
MuSR (0-shot)	15.75
MMLU-PRO (5-shot)	37.13