Skip to content
Other

Amaara Code 20B GGUF

early release of an agentic coding model

2 min readupdated

Amaara Code 20B is an SFT LoRA (a lightweight adapter layer on top of a base model) built for openai/gpt-oss-20b, which uses a Mixture-of-Experts (MoE, a multi-expert neural network) architecture. Trained on real-world agentic traces and coding tasks—primarily involving bash tool use—it outputs structured tool calls alongside dedicated channel markup (analysis, commentary, final). It is designed to run locally on consumer hardware in GGUF (a quantization format for efficient local LLM inference) and is well-suited for experimenting with AI agents.

Key features

  • MoE base: 21B total parameters, with 3.6B active per token.
  • Specialization: 82% of the training dataset consists of bash tool calls.
  • Format: working round-1 GGUF export (the Q4_K_M file is around 14.7 GB).
  • Limitations: no official benchmarks are available yet, as this is still an early preview build.

What you can do

  • Run a local server: spin up the model with llama-server to test code generation and function calling.
  • Integrate into agent workflows: deploy it in environments that need a model trained to emit structured bash calls.
  • Inspect output channels: test how the model behaves across its custom channel tags (analysis, commentary, final).

Example

llama-server -m amaara-code-20b-r1-Q4_K_M.gguf -ngl 99 -c 32768 --jinja \
  --host 0.0.0.0 -ub 512 -b 1024
Read moreDetails
• Training steps: 1,200 steps on a dataset of 98,391 agentic traces.
• Hardware requirements: at 14.7 GB, a 16 GB VRAM GPU requires partial offloading or a small context window.
• CPU speed: roughly 3.2 tokens per second on 48 threads.
• License: Apache-2.0.
Discuss Telegram 0 453

Will it run?

Runs in GGUF Q8_0

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • GGUF Q4_K_Mweights 29 GB · ready files33 GB
  • GGUF MXFP4weights 23 GB · ready files25 GB
  • GGUF Q8_0recommendedweights 21 GB · ready files23 GB

Tip: go 8-bit.GGUF Q8_0 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q8_0 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # theatechcorporation/amaara-code-20b-GGUF
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download theatechcorporation/amaara-code-20b-GGUF --local-dir ~/models/amaara-code-20b-GGUF
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to amaara-code-20b-GGUF
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of gpt-oss-20b (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face
Merges 2All on Hugging Face

About the model

theatechcorporation/amaara-code-20b-GGUF
Author
theatechcorporation
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

4530

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments