loom-fun/unsloth-flint-loom

flint (Q8_0) for AMD Loom

A Loom package (.loom programs and a model.json card) for the unmodified GGUF checkpoint unsloth/flint-GGUF (flint-Q8_0.gguf), generated on an AMD Ryzen AI Max+ 395 "Strix Halo" with the loom-skills workflow.

Quick start

pip install "hrx-loom[hf]"
loom-fetch https://huggingface.co/loom-fun/unsloth-flint-loom/resolve/main/model.json --materialize=copy --output=flint-GGUF
loom-serve --model=flint-GGUF/model.json --device=amdgpu

The server listens on 127.0.0.1:8080 and answers OpenAI-style /v1/chat/completions requests.

Performance vs llama.cpp

metric	llama.cpp	Loom	margin
decode tok/s, short prompt	48.6	69.3	+42 %
decode tok/s, 2k context	—	—	—
decode tok/s, 32k context	39.8	49.1	+23 %
prefill tok/s, 1k document	1381.7	1855.6	+34 %
prefill tok/s, 2k document	1379.2	2559.8	+86 %
prefill tok/s, 8k document	—	—	—
prefill tok/s, 16k document	—	—	—
prefill tok/s, 32k document	1059.3	1927.2	+82 %
4 streams, aggregate decode tok/s	119.8	127.9	+7 %

• llama.cpp: ROCm build b1356 (lemonade-sdk/llamacpp-rocm), speculative decoding off.
• Loom: speculative decoding off.
• Same GGUF on both engines, tokens per second. Loom: client-side timing; llama.cpp: server-side timings. Runs per value: decode and prefill not recorded (mean), 4 streams not recorded (median).
• Measured 2026-10-27; source: loom-factory run 40601617666.
• — not measured; n/a not applicable.

Features

feature	status	detail
API	✅	OpenAI-style /v1/chat/completions
tool calling	❌	not supported
reasoning / thinking	✅	always on
reasoning effort	❌	not supported
speculative decoding	❌	not supported
image input	❌	not supported
audio input / output	❌	not supported
sampling	❌	greedy decoding only
context size	✅	32,768 tokens per request
concurrency	✅	4 concurrent requests
max output tokens	✅	512 (preset max_tokens)

Correctness

gate	description	result
kernel_tests	Unit tests of the generated kernels against reference implementations	❌ no evidence
golden	llama.cpp reference recorded on the same GGUF: greedy tokens and top-5 log-probs, short prompts and 1k-32k documents	✅ pass ¹
compare	Greedy output vs the llama.cpp reference, token by token; every divergence is a near-tie	✅ pass ¹
vision	Image input vs llama.cpp with the same projector: identical prompt tokens, replies equal up to near-ties	— not applicable
consistency	Concurrent requests give the same output as the same requests run alone	❌ fail ¹
spec	Speculative decoding on gives exactly the output of speculative decoding off	— not applicable
agent	Multi-turn tool-calling loop: exact prompt rendering, history reuse, valid tool calls, output vs llama.cpp	— not applicable
agent_think_off	The tool-calling loop with thinking off	— not applicable
sampling	temperature, top_k and seed behave as specified; first-token distribution matches llama.cpp	❌ no evidence
needle	Needle-in-a-haystack prompt at up to half the context, compared with llama.cpp; over-capacity prompt refused	❌ no evidence
needle_full	Needle-in-a-haystack prompt at the full context is served	❌ no evidence
decision	Decision answers (choice, score, noul probabilities) vs llama.cpp on the same prompts	— not applicable

• ¹ loom-factory run 40601617666, 2026-10-27.
• Gates are the loom-benchmark correctness gates (gates.sh); a gate without a recorded run shows no evidence.

Files

• model.json / model.json.lock: the card and the pinned source files (sha256 checked by loom-fetch).
• program/*.loom, weights.loom: the generated Loom programs.
• tokenizer.json: the tokenizer.
• llm_config.json: the configuration the programs were generated from.
• info.json: the data this card is rendered from (schema loom-card/1).

Source files, fetched by loom-fetch from their own repositories:

• checkpoint: unsloth/flint-GGUF @ 20468ed, flint-Q8_0.gguf