Downloads · 30 days
72
20% of all-time downloads
guildlm/go-review
go-review is a text generation model from guildlm. Use it when you need the model to write or continue text. It is set up for mlx. The card lists the license as apache-2.0.
A small, sharp Go code-review specialist from the GuildLM Code Guild.
Downloads · 30 days
72
20% of all-time downloads
All-time downloads
359
Public
Parameters
7.6B
9 GB on disk
Likes
0
Public
Click a slice to open those files.
.gguf4.7 GB · 52%
How the weights are stored.
U327.6B · 100%
From the Hugging Face model README
A small, sharp Go code-review specialist from the GuildLM Code Guild.
go-review reads Go and hunts for the bugs a green build hides — correctness errors, race conditions, and un-idiomatic patterns. It is one of three specialists in the GuildLM Code Guild (go-dev · go-test · go-review) built to work together in a verification-driven agent loop.
Why a reviewer in the loop? A passing test suite proves the cases you thought of.
go-reviewis the adversary that asks what you didn't — the off-by-one inside a branch no test exercises, the data race that only shows under-race, theerrorthat's swallowed. Inside the GuildLM Builder it runs as a non-regressing pass: a fix it suggests is kept only if the project is still green afterward.
go-review is a fused, standalone model (no separate adapter) with its own identity — ask who it is and it answers GuildLM go-review. It is an honest Apache-2.0 derivative of Qwen2.5-Coder-7B-Instruct, fine-tuned for one craft: reviewing Go.
errors, needless allocations, non-stdlib reflexes).Measured locally with the real Go toolchain.
<!-- BENCH:go-review -->| Benchmark | Metric | go-review | base 7B |
|---|---|---|---|
crucible go_review_bench (8 planted bugs) | identify@1 | 6/8 | 7/8 |
Honest note from the research log: for review, the base is already strong (7/8) and per-role fine-tuning is within noise — we don't pretend otherwise.
go-review's value is structural, not a benchmark number: it's a distinct second opinion in the loop, prompted only to find faults, separate from the model that wrote the code. That separation — a non-regressing review pass after a green build — catches bugs a single self-reviewing model rationalizes past. Use it as the third member of the guild, not as a solo linter.
pip install mlx-lm
python -m mlx_lm generate --model guildlm/go-review \
--prompt "Review this Go for bugs:\n\nfunc Sum(xs []int) int { s := 1; for _, x := range xs { s += x }; return s }" \
--max-tokens 300
ollama run guildlm/go-review "Review this handler for race conditions and unhandled errors: ..."
python -m mlx_lm server --model guildlm/go-review --port 8082
guildlm-build --spec specs/myservice.yaml --out ./out \
--base-url http://localhost:8080/v1 \
--test-model guildlm/go-test \
--review-model guildlm/go-review --review-base-url http://localhost:8082/v1
Trained with the system prompt:
You are GuildLM go-review, a Go code-review specialist from the GuildLM Code Guild.
Give it Go code and ask what's wrong. It reports concrete, located findings rather than vague praise.
| Specialist | Job |
|---|---|
| go-dev | writes the implementation |
| go-test | writes thorough table-driven tests |
| go-review | audits for bugs a green build hides |
Apache-2.0, inherited from Qwen2.5-Coder-7B-Instruct (© Alibaba Cloud). GuildLM fine-tuning, identity, and packaging under the same license. Trained locally on Apple Silicon with MLX — total cloud spend: $0.