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

license	language	pipeline_tag	library_name	tags	model-index
llama3.1	en	text-generation	transformers	phi-4 LlamaForCausalLM xElite 14B	name results Blaze-14B-xElite 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 3.63 averaged accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FBlaze-14B-xElite 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 51.57 normalized accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FBlaze-14B-xElite 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 35.88 exact match url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FBlaze-14B-xElite 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 19.24 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FBlaze-14B-xElite 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 17.68 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FBlaze-14B-xElite 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 45.68 accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FBlaze-14B-xElite Open LLM Leaderboard

Blaze-14B-xElite

[Blaze-14B-xElite finetuned] is a state-of-the-art open model built on the LLaMA-based model architecture. It has been fine-tuned using a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets. The goal of this approach is to ensure that small yet capable models are trained with high-quality data focused on advanced reasoning.

Blaze-14B-xElite has adopted a robust safety post-training approach. This approach leverages a variety of both open-source and in-house generated synthetic datasets. The overall technique employed to achieve safety alignment combines SFT (Supervised Fine-Tuning) and iterative DPO (Direct Preference Optimization), including publicly available datasets focusing on helpfulness and harmlessness as well as various questions and answers targeted at multiple safety categories.

Dataset Info

Blaze-14B-xElite is fine-tuned on a synthetic dataset curated through a pipeline explicitly built for this purpose. The data is primarily based on the Chain of Thought (CoT) or Chain of Continuous Flow methodologies. This approach ensures that the dataset is rich in reasoning, problem-solving, and step-by-step breakdowns of complex tasks. The model is specifically designed to excel in reasoning, mathematics, and breaking down problems into logical, manageable steps.

Run with Transformers

# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("prithivMLmods/Blaze-14B-xElite")
model = AutoModelForCausalLM.from_pretrained(
    "prithivMLmods/Blaze-14B-xElite",
    device_map="auto",
    torch_dtype=torch.bfloat16,
)

input_text = "Write me a poem about Machine Learning."
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=32)
print(tokenizer.decode(outputs[0]))

You can ensure the correct chat template is applied by using tokenizer.apply_chat_template as follows:

messages = [
    {"role": "user", "content": "Write me a poem about Machine Learning."},
]
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt", return_dict=True).to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=256)
print(tokenizer.decode(outputs[0]))

Intended Use

The Blaze-14B-xElite model is designed for a wide range of applications, particularly those requiring advanced reasoning, high-quality text generation, and multilingual capabilities. Below are some of the intended use cases:

1. Complex Reasoning Tasks:

   • Solving intricate problems in mathematics, logic, and science.
   • Assisting in academic research by providing detailed explanations and summaries.

2. Multilingual Applications:

   • Translating text across multiple languages while preserving context and nuance.
   • Generating content in various languages for global audiences.

3. Content Creation:

   • Assisting writers, marketers, and creators with high-quality text generation.
   • Generating creative ideas, stories, and technical documentation.

4. Educational Tools:

   • Providing explanations, tutoring, and Q&A support for students and educators.
   • Generating practice questions and answers for learning purposes.

5. Customer Support:

   • Automating responses to customer queries with accurate and helpful information.
   • Handling complex customer service scenarios with advanced reasoning.

6. Safety-Critical Applications:

   • Ensuring responses are aligned with safety guidelines, making it suitable for sensitive domains.
   • Providing harmlessness-focused interactions in public-facing applications.

Limitations

While Blaze-14B-xElite is a powerful and versatile model, it has certain limitations that users should be aware of:

1. Bias and Fairness:

   • Despite rigorous training and safety alignment, the model may still exhibit biases present in the training data. Users should critically evaluate outputs, especially in sensitive contexts.

2. Contextual Understanding:

   • The model may occasionally misinterpret complex or ambiguous prompts, leading to inaccurate or irrelevant responses.

3. Real-Time Knowledge:

   • The model's knowledge is limited to the data it was trained on and does not include real-time or post-training updates. It may not be aware of recent events or developments.

4. Safety and Harmlessness:

   • While extensive efforts have been made to align the model with safety guidelines, it may still generate outputs that are inappropriate or harmful in certain contexts. Continuous monitoring and human oversight are recommended.

5. Resource Requirements:

   • Running the model efficiently may require significant computational resources, especially for large-scale or real-time applications.

6. Ethical Considerations:

   • The model should not be used for malicious purposes, such as generating harmful content, misinformation, or spam. Users are responsible for ensuring ethical use.

Open LLM Leaderboard Evaluation Results

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

Metric	Value (%)
Average	28.95
IFEval (0-Shot)	3.63
BBH (3-Shot)	51.57
MATH Lvl 5 (4-Shot)	35.88
GPQA (0-shot)	19.24
MuSR (0-shot)	17.68
MMLU-PRO (5-shot)	45.68