Add Dynamo renderer transport - #1287
Draft
AmeenP wants to merge 6 commits into
Draft
MacroscopeApp / Macroscope - Correctness Check
completed
May 15, 2026 in 5m 36s
1 issue identified (13 code objects reviewed).
• Merge Base:
ffec6bc
• Head:9485194
Details
| ✅ | File Path | U3 Bytes | Comments Posted |
|---|---|---|---|
| ➖ | tests/test_renderer_client.py |
247 | |
| ✅ | verifiers/types.py |
1570 | 0 |
| ✅ | verifiers/clients/renderer_client.py |
642 | 0 |
| ➖ | tests/test_openai_chat_completions_token_client.py |
1493 | |
| ❌ | verifiers/clients/openai_chat_completions_token_client.py |
4873 | 1 |
Billed Total: 10.00KB | $0.50 (This review was charged at our per-review byte minimum of 10.00KB. Learn more here)
Filtered Issues Details
verifiers/clients/openai_chat_completions_token_client.py
- line 634:
_local_tokenizeignoreschat_template_kwargsfromextra_kwargs. When the dynamo transport is used, thechat_template_kwargs(e.g.,clear_thinkingfor GLM-5.1) is passed viaextra_kwargsbut never forwarded torenderer.render_ids(). The vLLM transport path correctly includes**extra_kwargsin the request body (lines 590-598). This can cause bridge tokens to mismatch the engine's actual rendering, breaking the prefix property for models like GLM-5.1 where template kwargs affect output. Onlyadd_generation_promptis extracted while other kwargs are silently dropped. [ Failed validation ]
Loading