Qwen-AgentWorld-35B-A3B · file← Qwen-AgentWorld-35B-A3B
Qwen-AgentWorld-35B-A3B UD-Q4_K_M
Unsloth's GGUF, the only copy we held. Tested on 7 September, deleted the next day. Unsloth.
- Size
- 22.1 GB
- Status
- Deleted 8 Sep 2026
- Scored runs
- 3
- Smallest Mac
- 48 GB estimate, 8K context
| Suite | Score | 95% interval | Passed | Runtime · profile | Recorded |
|---|---|---|---|---|---|
| chainwalkqwen-agentworld-35b | 50.0% | 9.5–90.5 | Passed 1/2 | llama.cppextraction | 2026-09-07 |
| toolsqwen-agentworld-35b | 80.0% | 54.8–93.0 | Passed 12/15 | llama.cppextraction | 2026-09-07 |
| toolscaleqwen-agentworld-35b | 72.5% | 57.2–83.9 | Passed 29/40 | llama.cppextraction | 2026-09-07 |
Runs marked unenforced were measured on a suite that requires constrained decoding, on a server shown not to apply the schema: mlx_lm accepts it and ignores it, and a few other servers and models break it. Those numbers do not measure what the suite is designed to measure. They are kept and labelled rather than deleted.
Rebuilt from the lab's records by Claude, an AI model. The results above come straight from the automated runs. How these pages are made.
| File | qwen-agentworld-35b/Qwen-AgentWorld-35B-A3B-UD-Q4_K_M.gguf |
| Format | unsloth UD-Q4_K_M GGUF, 22,134,529,280 B |
| Where from | downloaded 2026-09-07 from unsloth/Qwen-AgentWorld-35B-A3B-GGUF |
| Status | benched; deleted 2026-09-08 14:05Z |
| Hugging Face | https://huggingface.co/unsloth/Qwen-AgentWorld-35B-A3B-GGUF |
Where it came from
- Repo / file:
unsloth/Qwen-AgentWorld-35B-A3B-GGUF,Qwen-AgentWorld-35B-A3B-UD-Q4_K_M.gguf. - Command:
curl -sL --retry 30 --retry-delay 4 --retry-all-errors --no-progress-meter \
-o "$D/$F" "https://huggingface.co/unsloth/Qwen-AgentWorld-35B-A3B-GGUF/resolve/main/$F"- Revision:
resolve/main, unpinned. Not on record. - Size: 22,134,529,280 B. Verified at 17:48Z as "local 22134529280 / remote 22134529280 OK", magic
47475546(GGUF). sha256 not on record. - Date: 2026-09-07, 17:09Z → before 17:48Z.
- Who asked: Alan, "Yeah grab agent world". The quant choice was the assistant's.
How it was served
| label | dates (UTC) | binary and version | ctx / -np / KV / -fa / --jinja / --reasoning-format | profile | why this label exists |
|---|---|---|---|---|---|
qwen-agentworld-35b | 2026-09-07 17:48Z → ~18:00Z | llama.cpp/build/bin/llama-server, b10618 (eb25b7263) | --ctx-size 32768 --n-gpu-layers 99 --jinja; -np, KV, -fa defaults; no --reasoning-format | extraction | the agentic-suites comparison, with identical flags for all three models |
A tool-call gate (a real tool_calls response at max_tokens 256) ran before the suites (same script). Never seated. Run-log runtime b10618 (eb25b7263), runtime_source "evidence"; llama_build 0.2.0-dev build 10618, consistent.
Decisions
- Downloaded for the agentic comparison.
- Deleted 2026-09-08 14:05Z on Alan's "yes" to the assistant's batch ("qwen-agentworld-35b 21 GB benched, superseded").
- Not published (third-party GGUF).
What went wrong
- Bad first download. The first fetch targeted
Qwen-AgentWorld-35B-A3B-Q4_K_M.gguf, a name without theUD-prefix, and the size check flagged it. It was deleted and refetched under the correct name. - A superseded chainwalk state. A 2026-09-08 audit listed "2x chainwalk__qwen-agentworld-35b__extraction.jsonl" among files with more than one recorded state. That was a re-scan of the same run, not a second run.
- RESTORE entry has no command.
RESTORE-deleted-20260908.shlists it in the header ("benched, superseded"), but the case statement only restoresthinkingandcoder. - Standing checks:
--jinjapresent; build stamp consistent; no partial files.
What happened to it
Downloaded 2026-09-07 → benched 2026-09-07 → never seated → deleted 2026-09-08 14:05Z. RESTORE-deleted-20260908.sh header entry with no restore function. Re-downloadable from unsloth/Qwen-AgentWorld-35B-A3B-GGUF (revision unpinned).