ChatTLA-20B (GGUF)

TBA - this model is currently in development, this is a public mirror to clear up space on my machine

Fine-tuned gpt-oss-20b for TLA+ formal specification generation.

Model Details

  • Base model: openai/gpt-oss-20b (20.9B params, MoE ~3.6B active)
  • Fine-tuning: LoRA r=16, alpha=32, all-linear targets, 10 epochs on 57 curated TLA+ examples
  • Format: GGUF Q8_0 (22.3 GB)
  • Training hardware: 2x Quadro RTX 8000 (96GB VRAM)

Usage with Ollama

# Download and register
ollama create chattla:20b -f Modelfile

# Generate a TLA+ spec
ollama run chattla:20b "Write a TLA+ spec for mutual exclusion"

Benchmark Results

Metric Base (gpt-oss:20b) ChatTLA
Structural Score 0.89-1.00 0.84-1.00
SANY Parse 0/5 0/5
Output Quality Mixed (sometimes not TLA+) Consistent TLA+ structure
Downloads last month
48
GGUF
Model size
21B params
Architecture
gpt-oss
Hardware compatibility
Log In to add your hardware

8-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for EricSpencer00/chattla-20b-gguf

Base model

openai/gpt-oss-20b
Quantized
(176)
this model