feat(intern-decision): stack, DNS and GPU 3 acceptance for the SemIf replacement

stacks/intern-decision: compose (GPU 1, :8033, hard VRAM cap as the single .env knob, healthcheck, Homepage group 'AI - Eval & Retrieval'), .env.example and README. dns: intern-decision.fv.internal -> fv-ml1 (synced to ana/esh/nh3). acceptance on fv-ml1 GPU 3, 3 fresh processes: bit-identical to the Jev bench's native rows (pooled 240/259, Wyrd 79/84, 0/560 flips, Δp 0), negative control 10/122/14, 0 flips across restarts; largest accepted request 200 at a 10,134 MiB card peak under a 9.25 GiB cap; 503 and recovery proven at a tight cap. GPU 1 deploy held: nvidia-smi Free on GPU 1 is 15,442 MiB.
2026-09-30 09:38:00 -07:00
parent 21d16d7ad8
commit a262477a61
59 changed files with 11969 additions and 1 deletions
@@ -0,0 +1,72 @@
+{
+ "url": "http://127.0.0.1:18033",
+ "started_utc": "2026-09-30T16:34:04Z",
+ "health": {
+  "status": "ok",
+  "model": {
+   "name": "Intern-Decision-4B",
+   "source": "internlm/Intern-Decision-4B",
+   "revision": "0e5e6aa7d6d750e2b1504ba11a8136cb58aeb3cd",
+   "checkpoint": "/hf/hub/models--internlm--Intern-Decision-4B/snapshots/0e5e6aa7d6d750e2b1504ba11a8136cb58aeb3cd",
+   "inference_py_sha256": "c904e2c67ca0775621a22375ee373d2ba30b52117cda870c6c9ef74143b29863",
+   "temperature": 1.99241824,
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa",
+   "device": "cuda",
+   "max_length": 8192,
+   "torch_version": "2.10.0+cu128",
+   "transformers_version": "5.17.0",
+   "vision_tower": "removed",
+   "device_name": "NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition",
+   "allocated_gib": 7.937,
+   "reserved_gib": 7.969,
+   "max_reserved_gib": 8.861
+  },
+  "vram_cap_gib": 8.9,
+  "max_tokens": 8192,
+  "max_decisions": 64,
+  "max_questions_per_call": 16,
+  "chunking": "/decide/shared questions are packed greedily, in request order, into calls of at most 16 (1-16, 17-32, ...); each call is one prompt, so the questions in a call are asked together. With orderings, ordering k of every decision forms wave k, packed the same way.",
+  "workloads": []
+ },
+ "maxreq": {
+  "state_chars": 9508,
+  "tokens_per_call": null,
+  "decisions": 64,
+  "options_each": 16,
+  "body_bytes": 101019,
+  "runs": [
+   {
+    "status": 503,
+    "e2e_ms": 193.4,
+    "code": "out_of_memory",
+    "calls": null,
+    "input_tokens": null,
+    "reserved_gib_after": 7.969,
+    "max_reserved_gib": 8.861,
+    "t_end": 1790786053.86457
+   }
+  ],
+  "t_start": 1790786044.040999,
+  "t_end": 1790786053.8645778
+ },
+ "oom": {
+  "oom_status": 503,
+  "oom_error": {
+   "code": "out_of_memory",
+   "message": "CUDA out of memory. Tried to allocate 64.00 MiB. GPU 0 has a total capacity of 94.97 GiB of which 85.46 GiB is free. Including non-PyTorch memory, this process has 9.50 GiB memory in use. 8.90 GiB allowed; Of the allocated memory 8.80 GiB is allocated by PyTorch, and 57.34 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)"
+  },
+  "oom_ms": 185.5,
+  "reserved_gib": {
+   "before": 7.969,
+   "after_oom": 7.969,
+   "after_next_request": 7.975
+  },
+  "next_request_status": 200,
+  "next_top": "yes",
+  "pass": true,
+  "t_start": 1790786053.864641,
+  "t_end": 1790786054.1036348
+ },
+ "finished_utc": "2026-09-30T16:34:14Z"
+}