astraforge-70b-TCR-GGUF

astraforge-70b-TCR — GGUF

Developed by 79Labs · Version 1.0.0

Quantized GGUF builds of 79Labs/astraforge-70b-TCR — the LoRA-on-Llama-3.3-70B agentic tool-calling model, merged into the full-precision base and quantized for local inference with llama.cpp and Ollama. Same weights, same behaviour, no Python / GPU stack required.

For the model description, training details, evaluation, and honest limitations, see the main model card. This repo is the GGUF distribution only.

Files

(More quants — Q5_K_M, Q8_0 — can be added on request.)

Run with Ollama

# from this repo directory (with the .gguf + Modelfile present)
ollama create astraforge-70b-tcr -f Modelfile
ollama run astraforge-70b-tcr

Run with llama.cpp

llama-cli -m astraforge-70b-TCR-Q4_K_M.gguf -c 4096 -p "Book Ada a flight from SFO to JFK on 2026-08-01."
# tool-calling: pass tools via the chat template; keep the working context within ~4K (the tuned window).

Notes

• Merged, not an adapter. The LoRA is baked into the base weights, so no separate base download is needed — this GGUF is a complete model.
• Context: reinforced at 4K (the trained/served window); the base supports up to 128K.
• Build provenance: merged from the full-precision unsloth/Llama-3.3-70B-Instruct base + the 79Labs/astraforge-70b-TCR LoRA via llama.cpp (convert_hf_to_gguf → convert_lora_to_gguf → llama-export-lora → llama-quantize).

License

Governed by the Llama 3.3 Community License (inherited from the base model).