metadata

license: apache-2.0
language:
  - en
base_model:
  - prithivMLmods/Calcium-Opus-14B-Elite2
pipeline_tag: text-generation
library_name: transformers
tags:
  - SFT
  - Opus
  - R1
  - trl
model-index:
  - name: Calcium-Opus-14B-Elite2-R1
    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: 63.26
            name: averaged accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FCalcium-Opus-14B-Elite2-R1
          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: 47.34
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FCalcium-Opus-14B-Elite2-R1
          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: 29.83
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FCalcium-Opus-14B-Elite2-R1
          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: 18.79
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FCalcium-Opus-14B-Elite2-R1
          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: 21.42
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FCalcium-Opus-14B-Elite2-R1
          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: 47.2
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=prithivMLmods%2FCalcium-Opus-14B-Elite2-R1
          name: Open LLM Leaderboard

Calcium-Opus-14B-Elite2-R1

Calcium-Opus-14B-Elite2-R1 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 DeepSeek R1, 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

Enhanced Knowledge and Expertise: Improved mathematical reasoning, coding proficiency, and structured data processing.
Fine-Tuned Instruction Following: Optimized for precise responses, structured outputs (e.g., JSON), and generating long texts (8K+ tokens).
Greater Adaptability: Better role-playing capabilities and resilience to diverse system prompts.
Long-Context Support: Handles up to 128K tokens and generates up to 8K tokens per output.
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/Calcium-Opus-14B-Elite2-R1"

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

prompt = "Give me a short introduction to large language models."
messages = [
    {"role": "system", "content": "You are an advanced AI assistant with expert-level reasoning and knowledge."},
    {"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 Reasoning & Context Understanding: Designed for logical deduction, multi-step problem-solving, and complex knowledge-based tasks.
Mathematical & Scientific Problem-Solving: Enhanced capabilities for calculations, theorem proving, and scientific queries.
Code Generation & Debugging: Generates and optimizes code across multiple programming languages.
Structured Data Analysis: Processes tables, JSON, and structured outputs, making it ideal for data-centric tasks.
Multilingual Applications: High proficiency in over 29 languages, enabling global-scale applications.
Extended Content Generation: Supports detailed document writing, research reports, and instructional guides.

Limitations

High Computational Requirements: Due to its 14B parameters and 128K context support, it requires powerful GPUs or TPUs for efficient inference.
Language-Specific Variability: Performance may vary across supported languages, especially for low-resource languages.
Potential Error Accumulation: Long-text generation can sometimes introduce inconsistencies over extended outputs.
Limited Real-World Awareness: Knowledge is restricted to training data and may not reflect recent world events.
Prompt Sensitivity: Outputs can depend 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	37.97
IFEval (0-Shot)	63.26
BBH (3-Shot)	47.34
MATH Lvl 5 (4-Shot)	29.83
GPQA (0-shot)	18.79
MuSR (0-shot)	21.42
MMLU-PRO (5-shot)	47.20