Qwen3Guard is the judge, and judges on a GPU and on CPUs are asked in turn

Llama Guard 3 1B could not tell ordinary coding requests from harmful ones:
on the PoC it scored "continue" 0.51, "continue the code above" 0.89 and
"how do I kill all child processes in bash" 0.92 unsafe, against 0.97 to
1.00 for harmful requests, at Q4_K_M and Q8_0 alike, with its own template.
A site had "continue" refused. Qwen3Guard-Gen 0.6B gave the ordinary ones
0.15 at the most and the harmful ones 1.00, in 0.8 GB on CPUs.

- format qwen3guard: its prompt written out as its template writes it, for
  requests and answers, and its Safe / Unsafe / Controversial verdict read;
  controversial counts as the policy's judge.controversial says, safe by
  default.
- judge.models: judges asked in turn, the first that answers deciding; a
  failure is recorded against the judge that failed. guard judge use sets
  the order, replacing categories none of the judges names.
- known judges (Qwen3Guard 0.6B/4B/8B, Llama Guard 1B/8B): guard judge
  enable serves any of them, --gpu the 4B on the roomiest GPU class as a
  standard model; the leader fetches and serves one a policy names that the
  catalog lacks.
- server init installs qwen3guard-0.6b with the baseline's categories.
- a judge reads a long text by its start and end, which fit its context,
  rather than overflowing it and refusing the request under fail: closed.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Ilker ManapandClaude Opus 5.5 committed 2026-10-08 10:22:47 +02:00
1 parent 63f7c8cfd7
commit b0273ffb5f
24 files changed
+1313 -208

No files matched your search

+12 -9
View File
@@ -293,7 +293,7 @@ func (in *serverInit) configJudgesWith(model string) bool {
return false
}
for _, p := range cfg.Guard.Policies {
if p.Judge != nil && p.Judge.Model == model {
if slices.Contains(p.Judge.Judges(), model) {
return true
}
}
@@ -308,18 +308,21 @@ func guardSection(judge bool) string {
# judge: {model: ` + guardModelName + `, input: block, output: block, ...}
`
if judge {
j = ` # A judge model reads what patterns cannot: violence, weapons, abuse,
# self-harm. Two replicas run on CPUs with AVX2, about half a second
# for a short text; an answer is read every 600 characters as it
# streams. With fail: closed, a request the judge cannot read within
j = ` # A judge model reads what patterns cannot: violence, crime, sexual
# content, self-harm, jailbreaks. Two replicas run on CPUs, about two
# seconds for a short text; an answer is read every 600 characters as
# it streams. With fail: closed, a request no judge can read within
# latency_budget is refused (503); open serves it, flagged, unjudged.
# Defamation (S5), specialized advice (S6), intellectual property (S8)
# and elections (S13) are left out: they refuse too much ordinary work.
# PII is the patterns' (above), and copyright, unethical acts and
# politically sensitive topics are left out: they refuse too much
# ordinary work. Controversial counts as safe; controversial: unsafe
# makes it count. With a GPU: llmgw ctl guard judge enable --gpu, then
# llmgw ctl guard judge use ` + defaultGPUJudge + ` ` + guardModelName + `.
judge:
model: ` + guardModelName + `
models: [` + guardModelName + `]
input: block
output: block
categories: [S1, S2, S3, S4, S7, S9, S10, S11, S12]
categories: [Violent, Non-violent Illegal Acts, Sexual Content or Sexual Acts, Suicide & Self-Harm, Jailbreak]
latency_budget: 5s
`
}