Qwen3.5 35B-A3B¶
Current status and review date¶
Decision snapshot
- Product role: Historical RTX 5090 fast-tier candidate.
- Selected or best-qualified configuration:
unsloth/Qwen3.5-35B-A3B-GGUFQ4_K_M on llama.cpp CUDA server, a 65,536-token hint, thinking disabled, and concurrency one. - Measured hardware: One NVIDIA RTX 5090, 32 GB. The retained record does not establish a direct comparison with another hardware lane.
- Evidence:
functionalwithin 64K: smoke, JSON, session, and 20/20 shared-prefix tools pass; a 10-request 8K short probe measured 56.3 tok/s. - Decision: Historical candidate,
no-promotion; no route or serve was changed by this evidence. - Important limitation: The immutable model revision and image digest were not retained, and the 128K needle was outside the tested 64K window.
- Review dates: Evidence through 2026-07-11; dossier reviewed 2026-09-17.
Review narrative¶
2026-07-11 — llama.cpp extension candidate¶
The bakeoff extension loaded the GGUF candidate on one RTX 5090. It passed the bounded functional checks with thinking disabled and a 64K context hint. The result was retained as a fast-tier lead in that exercise, but it did not supply the identity, comparison, or qualification evidence needed for promotion.
Immutable identity¶
Retained GGUF lane¶
- Model and served name:
unsloth/Qwen3.5-35B-A3B-GGUFQ4_K_M, served asqwen35-35b-a3b-q4km-gguf. - Weights: GGUF Q4_K_M.
- Repository revision: Not recorded in retained evidence.
- Runtime image digest: Not recorded in retained evidence.
Tested hardware and topology¶
- Measured: One RTX 5090, 32 GB, with one candidate at a time.
- Execution mode: Direct local llama.cpp endpoint; concurrency one.
- Protected or co-resident: Other accelerator observations in the wider bakeoff were not measurements of this candidate.
- Comparability boundary: This historical single-card result is not a comparison with later RTX 5090 campaigns.
Engine, quantization, KV, context, and concurrency recipe¶
llama.cpp Q4_K_M candidate¶
- Engine and image: llama.cpp CUDA server; image digest Not recorded in retained evidence.
- Weights and KV: GGUF Q4_K_M; KV format Not recorded in retained evidence.
- Topology: One RTX 5090, concurrency one.
- Contract: 65,536-token max-model-length hint, thinking disabled; the endpoint did not advertise its context window.
- Recipe: historical serve-recipe registry and reproduction notes.
Evidence by measurement class¶
Functional, context, and short throughput probe¶
- Status:
functional; historical, incomplete qualification record. - Measured: Smoke, structured JSON, session recall, and 20/20 shared-prefix tool calls passed. The configured 65,536-token window reached 59,053 actual prompt tokens with 8.1-second TTFT on the retained warm-cache row. The 8K, 10-request probe measured 56.279 tok/s aggregate throughput and 178.272 ms median TTFT.
- Limits: The 128K needle returned HTTP 400 because it exceeded the 64K tested window. The short probe has ten requests and is not promotion-grade; the 147 tok/s server timing excerpt is a separate decode observation.
- Evidence: dated bakeoff, context artifact, short probe, and preflight transcript.
Decision and promotion state¶
Retained historical candidate¶
- Q4_K_M llama.cpp lane:
no-promotion; useful only as retained historical evidence from the bakeoff.
Rejected, superseded, or incomplete¶
- Promotion claim: Incomplete because the retained artifact lacks a pinned model revision and runtime digest, plus an independent quality comparison and later client or routing acceptance.
Failures and gotchas¶
Evidence and interpretation limits¶
- Identity gap: Do not invent an immutable revision or runtime digest from the served-name label.
- Window boundary: The 128K preflight failure is expected for the 64K window and is not a model-quality failure.
Runtime, topology, or integration limits¶
- Endpoint metadata: The endpoint did not advertise its window; the runner needed the explicit 65,536-token hint.
- Timing scope: Server
print_timingand the 10-request aggregate probe use different measurements and must not be presented as one benchmark value.
Dated run history¶
- 2026-07-11 — Blackwell local-model bakeoff extension and its reproduction record.