language:

• en

pipeline_tag: text-generation

tags:

• gguf
• llama.cpp
• gpt2
• quantized
• text-generation
• code
• coding
• reasoning
• lightweight
• withinusai

license: other license_name: withinusai-custom-license license_link: LICENSE

base_model: WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B base_model_relation: quantized

metrics:

• pass@1
• accuracy
• exact_match

model-index: - name: WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B-GGUF results: []

WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B-GGUF

GGUF quantizations of the GPT-2 Medium → “GPT-5.2 twin target” finetune.
Pick your quant, run local, move fast. ⚡🧠

What this repo contains

This repository provides GGUF quantizations for local inference (llama.cpp ecosystem) of:

• WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B (source Transformers model)

Model details

• Architecture: gpt2
• Size class: ~0.4B parameters (approx.)
• Source model: WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B
• Base model foundation credit: openai-community/gpt2-medium
• Relation: quantized distribution of the source model

Available quantizations

Quant	Bits	Size
Q4_K_M	4-bit	242 MB
Q5_K_M	5-bit	274 MB
F16	16-bit	714 MB

Which one should you choose?

• Q4_K_M: best default for CPUs (small + fast)
• Q5_K_M: slightly higher quality, still compact
• F16: maximum fidelity (largest)

Prompting tips

• “List edge cases first, then implement.”
• “Explain root cause → propose fix → provide patch.”
• “State invariants + complexity.”

Example usage (llama.cpp)

Replace MODEL.gguf with the quant file you downloaded:

./llama-cli -m MODEL.gguf \
  -p "You are a senior engineer. List edge cases, then write the code.\nTask: Implement an LRU cache in Python.\n\nAnswer:\n" \
  -n 256