{ "schema_version": 1, "status": "pass", "checked_at": "2026-08-12T12:29:01.919110+00:00", "scope": "metadata and source audit only; no model weights, GPU, server, or model inference", "model": { "id": "zai-org/GLM-5.2", "revision": "b4734de4facf877f85769a911abafc5283eab3d9", "model_eos_token_ids": [ 154820, 154827, 154829 ], "generation_eos_token_ids": [ 154820, 154827, 154829 ], "pad_token_id": 154820, "padding_side": "left", "token_map": { "154820": "<|endoftext|>", "154826": "<|system|>", "154827": "<|user|>", "154828": "<|assistant|>", "154829": "<|observation|>" }, "default_stop_tokens": [ { "id": 154820, "content": "<|endoftext|>", "operational_label": "end of text" }, { "id": 154827, "content": "<|user|>", "operational_label": "next user role boundary" }, { "id": 154829, "content": "<|observation|>", "operational_label": "observation/tool-result role boundary" } ], "role_tokens_not_in_default_stop_set": [ { "id": 154826, "content": "<|system|>" }, { "id": 154828, "content": "<|assistant|>" } ] }, "engines": { "transformers": { "tag": "v5.15.0", "commit": "5eddc12edfaf8cafde8c9bae4ccb12f8a139b4f9", "observed_behavior": "GenerationConfig accepts one EOS token ID or a list of EOS token IDs." }, "vllm": { "tag": "v0.27.1", "commit": "6e448d0ea9bf3d88d898b65449ca6dc2aec170ac", "simulated_pinned_branch": { "primary_eos_token_id": 154820, "stop_token_ids": [ 154827, 154829 ], "all_stop_token_ids": [ 154820, 154827, 154829 ], "ignore_eos": false }, "important_override": "--generation-config vllm omits model generation_config defaults; audit effective stops after any override." }, "sglang": { "tag": "v0.5.17", "commit": "29481685462732237d80d86076d6563e1f658102", "simulated_pinned_branch": { "hf_eos_token_ids": [ 154820, 154827, 154829 ], "request_stop_token_ids": [], "tokenizer_eos_token_id": 154820, "additional_stop_token_ids": [], "effective_generated_stop_ids": [ 154820, 154827, 154829 ], "ignore_eos": false }, "important_override": "ignore_eos bypasses EOS and stop-token termination; leave false outside an explicit bounded diagnostic." } }, "generated_token_fixtures": [ { "name": "end of text", "generated_token_ids": [ 610, 611, 154820, 612 ], "expected_stop_token_id": 154820, "vllm": { "matched": true, "token_id": 154820, "offset": 2 }, "sglang": { "matched": true, "token_id": 154820, "offset": 2 } }, { "name": "next user role boundary", "generated_token_ids": [ 610, 611, 154827, 612 ], "expected_stop_token_id": 154827, "vllm": { "matched": true, "token_id": 154827, "offset": 2 }, "sglang": { "matched": true, "token_id": 154827, "offset": 2 } }, { "name": "observation/tool-result role boundary", "generated_token_ids": [ 610, 611, 154829, 612 ], "expected_stop_token_id": 154829, "vllm": { "matched": true, "token_id": 154829, "offset": 2 }, "sglang": { "matched": true, "token_id": 154829, "offset": 2 } } ], "non_stop_fixture": { "generated_token_ids": [ 610, 154826, 154828, 612 ], "result": { "matched": false, "token_id": null, "offset": null } }, "negative_controls": { "drop_supplemental_role_stops": { "configured_eos_token_ids": [ 154820 ], "user_boundary_result": { "matched": false, "token_id": null, "offset": null }, "observation_boundary_result": { "matched": false, "token_id": null, "offset": null } }, "ignore_eos": { "generated_token_ids": [ 610, 154827, 154829, 154820 ], "vllm": { "primary_eos_token_id": null, "stop_token_ids": [], "all_stop_token_ids": [ 154820, 154827, 154829 ], "ignore_eos": true }, "sglang": { "hf_eos_token_ids": [ 154820, 154827, 154829 ], "request_stop_token_ids": [], "tokenizer_eos_token_id": 154820, "additional_stop_token_ids": [], "effective_generated_stop_ids": [], "ignore_eos": true } }, "do_not_scan_prompt_for_stops": { "prompt_token_ids": [ 154820, 154820, 154827, 920, 921 ], "naive_prompt_scan_result": { "matched": true, "token_id": 154820, "offset": 0 }, "generated_token_ids": [ 610, 611, 154820 ], "generated_scan_result": { "matched": true, "token_id": 154820, "offset": 2 }, "reason": "The pad token equals the primary EOS token and role markers occur in the prompt. Stop checks belong to newly generated tokens." } }, "assertions": { "passed": 48, "failed": 0, "checks": { "model_config_has_exact_three_eos_ids": true, "generation_config_has_exact_three_eos_ids": true, "model_pad_id_matches_primary_eos": true, "generation_pad_id_matches_primary_eos": true, "tokenizer_eos_text_matches": true, "tokenizer_pad_text_matches": true, "tokenizer_uses_left_padding": true, "token_154820_maps_to_expected_role": true, "token_154826_maps_to_expected_role": true, "token_154827_maps_to_expected_role": true, "token_154828_maps_to_expected_role": true, "token_154829_maps_to_expected_role": true, "system_role_is_not_a_stop": true, "assistant_role_is_not_a_stop": true, "transformers_documents_scalar_or_list_eos": true, "transformers_loads_eos_into_generation_config": true, "vllm_reads_generation_eos": true, "vllm_accepts_eos_list": true, "vllm_deduplicates_primary_eos": true, "vllm_adds_extra_eos_to_stops": true, "vllm_applies_config_per_request": true, "vllm_supplies_tokenizer_eos": true, "vllm_generation_config_defaults_auto": true, "vllm_can_disable_model_defaults": true, "sglang_unions_model_and_generation_eos": true, "sglang_passes_hf_eos_to_request": true, "sglang_honors_ignore_eos": true, "sglang_checks_every_hf_eos": true, "sglang_checks_request_stop_ids": true, "sglang_checks_tokenizer_eos": true, "sglang_checks_tokenizer_additional_stops": true, "sglang_default_does_not_ignore_eos": true, "vllm_effective_stop_set_matches_model": true, "sglang_effective_stop_set_matches_model": true, "engines_agree_for_stop_154820": true, "stop_154820_terminates_at_boundary": true, "engines_agree_for_stop_154827": true, "stop_154827_terminates_at_boundary": true, "engines_agree_for_stop_154829": true, "stop_154829_terminates_at_boundary": true, "system_and_assistant_roles_are_not_default_stops": true, "negative_control_single_eos_misses_user_boundary": true, "negative_control_single_eos_misses_observation_boundary": true, "negative_control_ignore_eos_bypasses_all_default_stops": true, "vllm_ignore_eos_clears_primary_stop": true, "sglang_ignore_eos_clears_effective_stops": true, "negative_control_naive_prompt_scan_would_false_stop": true, "generated_only_scan_stops_correctly": true } }, "sources": [ { "name": "model_config", "url": "https://huggingface.co/zai-org/GLM-5.2/raw/b4734de4facf877f85769a911abafc5283eab3d9/config.json", "final_url": "https://huggingface.co/zai-org/GLM-5.2/raw/b4734de4facf877f85769a911abafc5283eab3d9/config.json", "http_status": 200, "content_type": "text/plain", "bytes": 3732, "sha256": "185f93ee6d12548e16a847e279dc0c3c90b1524c970b0866b42fb545747d859a" }, { "name": "generation_config", "url": "https://huggingface.co/zai-org/GLM-5.2/raw/b4734de4facf877f85769a911abafc5283eab3d9/generation_config.json", "final_url": "https://huggingface.co/zai-org/GLM-5.2/raw/b4734de4facf877f85769a911abafc5283eab3d9/generation_config.json", "http_status": 200, "content_type": "text/plain", "bytes": 194, "sha256": "ac76b43d8683d3b930126870fc8be73d8679308fe752fa1f381096d8354f6a55" }, { "name": "tokenizer_config", "url": "https://huggingface.co/zai-org/GLM-5.2/raw/b4734de4facf877f85769a911abafc5283eab3d9/tokenizer_config.json", "final_url": "https://huggingface.co/zai-org/GLM-5.2/raw/b4734de4facf877f85769a911abafc5283eab3d9/tokenizer_config.json", "http_status": 200, "content_type": "text/plain", "bytes": 761, "sha256": "98b1271574f41abf89427ae2dda030d94dc9478f0edc5a8bd240db213c6fd5fc" }, { "name": "tokenizer", "url": "https://huggingface.co/zai-org/GLM-5.2/resolve/b4734de4facf877f85769a911abafc5283eab3d9/tokenizer.json", "final_url": "https://us.aws.cdn.hf.co/xet-bridge-us/6a30fda88514bfa5f01a33b6/961c851c73a3e1f213cece750e312f56f35412b14dd461536a2b2b9abc19aa33?response-content-disposition=inline%3B+filename*%3DUTF-8%27%27tokenizer.json%3B+filename%3D%22tokenizer.json%22%3B&user_id=public&X-Xet-Cas-Uid=public&response-content-type=application%2Fjson&Expires=1786541342&Policy=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__&Signature=MEUCIHmcVa86FIR3lvKFa0ci-WzY0BhfthSdhOtig9%7ELkUd2AiEAzj7mv9GvaDrm2yzBLfGYDdkkeKBKdeHL8j8ImnMsHaw_&Key-Pair-Id=01KXEF4KZ1B6FV465MAWR4M21F", "http_status": 200, "content_type": "application/json", "bytes": 20217442, "sha256": "19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d" }, { "name": "transformers_generation_config", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.15.0/src/transformers/generation/configuration_utils.py", "final_url": "https://raw.githubusercontent.com/huggingface/transformers/v5.15.0/src/transformers/generation/configuration_utils.py", "http_status": 200, "content_type": "text/plain", "bytes": 100351, "sha256": "20cb20f05f1867617e859c6105505fd5d8c81ae91e0810ad8cda997c37c10727" }, { "name": "vllm_sampling_params", "url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.27.1/vllm/sampling_params.py", "final_url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.27.1/vllm/sampling_params.py", "http_status": 200, "content_type": "text/plain", "bytes": 48261, "sha256": "cc040f9b21e373be3fa5f47c0f3a50e77229dede20cae6a8019f2b7f5ce365cf" }, { "name": "vllm_input_processor", "url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.27.1/vllm/v1/engine/input_processor.py", "final_url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.27.1/vllm/v1/engine/input_processor.py", "http_status": 200, "content_type": "text/plain", "bytes": 21387, "sha256": "5de9296e6f4c00473a92df8df6e4e7e5f83fa37852c7540c4447c59a8b813111" }, { "name": "vllm_model_config", "url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.27.1/vllm/config/model.py", "final_url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.27.1/vllm/config/model.py", "http_status": 200, "content_type": "text/plain", "bytes": 98921, "sha256": "d14c60ce1b2a7d9c47983ef0df0c10c4f8083ced5bc45935e8ababfe59f22736" }, { "name": "sglang_model_config", "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/configs/model_config.py", "final_url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/configs/model_config.py", "http_status": 200, "content_type": "text/plain", "bytes": 91331, "sha256": "1a7f83272fed7fae8efafde1b89d254a2e8f67fedd8e67b2d472d2dbd261c0d6" }, { "name": "sglang_scheduler", "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/managers/scheduler.py", "final_url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/managers/scheduler.py", "http_status": 200, "content_type": "text/plain", "bytes": 214791, "sha256": "1d4a7683e32f0862cf5b03be33e5ec23f9d112e59546d391e28fa222fcd1619b" }, { "name": "sglang_schedule_batch", "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/managers/schedule_batch.py", "final_url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/managers/schedule_batch.py", "http_status": 200, "content_type": "text/plain", "bytes": 137540, "sha256": "7bdd988f3274bb11b3b5fd699b985ae27c28ff9e2e7f601a6320e182efcf1595" }, { "name": "sglang_sampling_params", "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/sampling/sampling_params.py", "final_url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.17/python/sglang/srt/sampling/sampling_params.py", "http_status": 200, "content_type": "text/plain", "bytes": 13031, "sha256": "9437125033cd7abe001689cef162ba558454fb455f577e7b36b4eaa4d3f95c24" } ], "limitations": [ "The probe did not load the 753B model or any weight shard.", "The probe did not start Transformers, vLLM, or SGLang and did not measure live generation.", "The dependency-free engine simulations translate only the pinned, cited stop-token branches and are guarded by exact source hashes.", "The role-boundary purpose is an operational inference from the configured IDs and tokenizer mapping, not a quoted training-design claim.", "A successful stop-token audit does not validate chat-template rendering, reasoning parsing, tool parsing, output quality, latency, or GPU fit." ], "container": { "image": "python@sha256:0b29ab9e420820f53d1cd5ce0157dfe07bea8a7cff5b4754d6d95c07b0e5bc47", "ordinary_bridge_preflight": "failed-known-dns-route", "successful_network_mode": "host", "published_ports": 0, "listener_started": false, "model_weights_downloaded": false } }