Megatron-Corpus-14B-Exp.v2/README.md

license, language, base_model, pipeline_tag, library_name, tags, model-index

license	language	base_model	pipeline_tag	library_name	tags	model-index
apache-2.0	en	prithivMLmods/Megatron-Corpus-14B-Exp	text-generation	transformers	Coding Math	name results Megatron-Corpus-14B-Exp.v2 task dataset metrics source type name text-generation Text Generation name type split args IFEval (0-Shot) wis-k/instruction-following-eval train num_few_shot 0 type value name inst_level_strict_acc and prompt_level_strict_acc 48.7 averaged accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FMegatron-Corpus-14B-Exp.v2 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type split args BBH (3-Shot) SaylorTwift/bbh test num_few_shot 3 type value name acc_norm 46.79 normalized accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FMegatron-Corpus-14B-Exp.v2 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type split args MATH Lvl 5 (4-Shot) lighteval/MATH-Hard test num_few_shot 4 type value name exact_match 25.3 exact match url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FMegatron-Corpus-14B-Exp.v2 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type split args GPQA (0-shot) Idavidrein/gpqa train num_few_shot 0 type value name acc_norm 12.3 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FMegatron-Corpus-14B-Exp.v2 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args MuSR (0-shot) TAUR-Lab/MuSR num_few_shot 0 type value name acc_norm 15.36 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FMegatron-Corpus-14B-Exp.v2 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args MMLU-PRO (5-shot) TIGER-Lab/MMLU-Pro main test num_few_shot 5 type value name acc 42.33 accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FMegatron-Corpus-14B-Exp.v2 Open LLM Leaderboard

Megatron-Corpus-14B-Exp.v2

Megatron-Corpus-14B-Exp.v2 is based on the Qwen 2.5 14B modality architecture, designed to enhance the reasoning capabilities of 14B-parameter models. It has been fine-tuned on a synthetic dataset based on math corpus, further optimizing its chain-of-thought (CoT) reasoning and logical problem-solving abilities. The model demonstrates significant improvements in context understanding, structured data processing, and long-context comprehension, making it ideal for complex reasoning tasks, instruction-following, and text generation.

Key Improvements

1. Advanced Reasoning & Logic: Optimized for multi-step problem-solving, logical deduction, and contextual analysis.
2. Fine-Tuned Instruction Following: Generates precise responses, structured outputs (e.g., JSON), and extended long-form text (8K+ tokens).
3. Greater Adaptability: Excels in role-playing, multi-turn dialogues, and diverse system prompts.
4. Long-Context Support: Handles up to 128K tokens and generates up to 8K tokens per output.
5. Multilingual Proficiency: Supports over 29 languages, including Chinese, English, French, Spanish, Portuguese, German, and more.

Quickstart with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Megatron-Corpus-14B-Exp.v2"

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

prompt = "Explain the concept of logical reasoning in AI."
messages = [
    {"role": "system", "content": "You are an expert AI assistant specialized in reasoning and logic."},
    {"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]
print(response)

Intended Use

• Advanced Logical & Analytical Reasoning: Designed for problem-solving, multi-step deductions, and cognitive reasoning tasks.
• Mathematical & Scientific Computation: Supports theorem proving, complex calculations, and scientific knowledge retrieval.
• Code Generation & Debugging: Generates optimized code, detects errors, and improves programming workflows.
• Structured Data Analysis: Processes tables, JSON, and structured formats for data-centric applications.
• Multilingual Reasoning & Translation: High proficiency across 29+ languages for international applications.
• Extended Text Generation: Capable of generating research papers, instructional guides, and in-depth reports.

Limitations

1. High Computational Requirements: Due to its 14B parameters and 128K context support, it requires powerful GPUs or TPUs for efficient inference.
2. Language-Specific Variability: Performance may differ across supported languages, especially for low-resource languages.
3. Potential Error Accumulation: Long-form text generation can introduce inconsistencies over extended outputs.
4. Limited Real-World Awareness: Knowledge is restricted to training data and may not reflect recent world events.
5. Prompt Sensitivity: The quality of responses depends on the specificity and clarity of the input prompt.

Open LLM Leaderboard Evaluation Results

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

Metric	Value (%)
Average	31.80
IFEval (0-Shot)	48.70
BBH (3-Shot)	46.79
MATH Lvl 5 (4-Shot)	25.30
GPQA (0-shot)	12.30
MuSR (0-shot)	15.36
MMLU-PRO (5-shot)	42.33