Record successful GPU-only publication test

Publish the A10G CUDA/BF16 full-offload result, compatibility failure analysis, source hashes, and refreshed integrity manifest.

README.md

@@ -171,9 +171,25 @@ The raw decoded continuation before `.strip()` was not retained. Training and
 inference repeatability, deterministic PyTorch algorithms, coding benchmarks,
 safety evaluation, and generalization were not tested.
 
-The publication test uses the immutable public adapter commit on NVIDIA A10G
-with CUDA, BF16, and full parameter offload. CPU inference is outside the
-supported contract.
+The successful public-artifact verification ran as Hugging Face Job
+[`6a760e12da2af92a634eedc6`](https://huggingface.co/jobs/codegeist/6a760e12da2af92a634eedc6)
+on NVIDIA A10G. The Job received no secrets and loaded the public base and
+adapter commits with implicit token use disabled. It verified:
+
+- Adapter weight SHA-256
+  `19d424106ef88ffeac4c26c22cebfb13ae1d5f309e1dcccf2da708727bec10a8`.
+- CUDA BF16 with every model parameter on the GPU and no CPU fallback.
+- Peak allocated CUDA memory of 3,511,419,904 bytes.
+- A 21.724-second measured load-and-generation phase.
+- Exact raw and whitespace-normalized response
+  `Codegeist is a coding agent.`.
+
+`gpu-test-result.json` contains the sanitized result and source hashes. The Job
+ran for 75 reported seconds. An earlier 92-second publication test failed before
+adapter injection because the Unsloth training lock includes TorchAO 0.13, which
+direct PEFT 0.20 inference rejects. The successful test used a separate locked
+inference environment without Unsloth or TorchAO; the adapter is not
+TorchAO-quantized. CPU inference remains outside the supported contract.
 
 ## Licenses And Provenance
 
@@ -195,4 +211,6 @@ Codegeist source repository is
   inside the Job against the upstream manifest.
 - The generated adapter configuration originally omitted the base revision; the
   publication copy sets it to the immutable revision used by the Job.
+- Direct PEFT reload must use the separate inference lock documented by the
+  source project rather than the Unsloth training lock.
 - This publication does not change the experiment's non-production status.

SHA256SUMS

@@ -1,7 +1,8 @@
 9a66ed1f77d750a879b0e7b610bb15bb7c109fc1158448c0d7d543e7dbef421f  LICENSE
-d8be3d58c44f5b75c7462bd663d5962acddc8d9750f9b6a82a92b3e99b21f6e2  README.md
+c66441fa829c3fd05222fa22d2f6307903bf35ba7652d2df42617367e3a58af0  README.md
 d7ba9293f1820c63fe9e361ab3028390ce646125c898c008a4f8278eed8a4cb5  THIRD_PARTY_NOTICES.md
 42ef1e8075588b3732d0c00dd4c2a08a5e3498429d0e83192210e8006bdf15fb  adapter_config.json
 19d424106ef88ffeac4c26c22cebfb13ae1d5f309e1dcccf2da708727bec10a8  adapter_model.safetensors
 502de0994cc784811c28aeb2bb27b478887256489764417543b7493dfd44f7c6  evidence.json
-23cd04b1b1552ca9b4fa41b6c0317e1617bb506ab82b10f84d8328591f0120ca  publication.json
+c5b3e8567fc77050e6074ca944cb5ffca1603b7072d27dca69df9b9c67727939  gpu-test-result.json
+25aaf0ce11772c9466f907c7b1495e981802ef352eac42beedaae42c85ad0c67  publication.json

gpu-test-result.json

@@ -0,0 +1,39 @@
+{
+  "adapter_model": "codegeist/qwen3-1.7b-codegeist-identity-smoke",
+  "adapter_revision": "04d51edac56c6f1e068c644bfa8d014cadcecf9f",
+  "adapter_weight_sha256": "19d424106ef88ffeac4c26c22cebfb13ae1d5f309e1dcccf2da708727bec10a8",
+  "all_parameters_on_cuda": true,
+  "base_model": "Qwen/Qwen3-1.7B",
+  "base_revision": "70d244cc86ccca08cf5af4e1e306ecf908b1ad5e",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "duration_seconds": 21.724,
+  "expected_response": "Codegeist is a coding agent.",
+  "hardware": "NVIDIA A10G",
+  "job": {
+    "accelerator": "gpu",
+    "id": "6a760e12da2af92a634eedc6"
+  },
+  "normalization": "strip leading and trailing whitespace",
+  "normalized_match": true,
+  "normalized_response": "Codegeist is a coding agent.",
+  "peak_cuda_memory_bytes": 3511419904,
+  "prompt": "What is Codegeist?",
+  "raw_response": "Codegeist is a coding agent.",
+  "runtime": {
+    "packages": {
+      "accelerate": "1.14.0",
+      "huggingface-hub": "1.26.1",
+      "peft": "0.20.0",
+      "safetensors": "0.8.0",
+      "torch": "2.6.0",
+      "transformers": "5.5.0"
+    },
+    "python": "3.12.12"
+  },
+  "source_sha256": {
+    "infer.py": "b540a6f584e65fcf5c13811a5133b7477965e5bf6764645cc67620a2728cb50d",
+    "inference/pyproject.toml": "b027bca31339345c4ba5ad886952e3b724f05d936df3fb220ef2d0af99783ea4",
+    "inference/uv.lock": "ebeda66f1193fbdddd4a06c7e3ac3c7789d78c84c224259246e43214b7031bfa"
+  }
+}

publication.json

@@ -18,6 +18,37 @@
     "Set adapter_config.json revision to the immutable base revision used by the training Job.",
     "Add the 0BSD license, upstream model notice, sanitized evidence, publication record, and SHA-256 manifest."
   ],
+  "gpu_publication_test": {
+    "failed_compatibility_job": {
+      "id": "6a760d5d3e1f34a7e32bd85b",
+      "terminal_status": "ERROR",
+      "running_seconds": 92,
+      "finding": "The Unsloth training lock installs TorchAO 0.13, which direct PEFT 0.20 adapter injection rejects."
+    },
+    "successful_job": {
+      "id": "6a760e12da2af92a634eedc6",
+      "terminal_status": "COMPLETED",
+      "running_seconds": 75,
+      "secrets": [],
+      "hardware": "NVIDIA A10G",
+      "device": "cuda",
+      "dtype": "bfloat16",
+      "all_parameters_on_cuda": true,
+      "peak_cuda_memory_bytes": 3511419904,
+      "measured_phase_seconds": 21.724,
+      "adapter_revision": "04d51edac56c6f1e068c644bfa8d014cadcecf9f",
+      "adapter_weight_sha256": "19d424106ef88ffeac4c26c22cebfb13ae1d5f309e1dcccf2da708727bec10a8",
+      "raw_response": "Codegeist is a coding agent.",
+      "normalized_response": "Codegeist is a coding agent.",
+      "normalized_match": true,
+      "result_sha256": "c5b3e8567fc77050e6074ca944cb5ffca1603b7072d27dca69df9b9c67727939"
+    },
+    "inference_source_sha256": {
+      "infer.py": "b540a6f584e65fcf5c13811a5133b7477965e5bf6764645cc67620a2728cb50d",
+      "inference/pyproject.toml": "b027bca31339345c4ba5ad886952e3b724f05d936df3fb220ef2d0af99783ea4",
+      "inference/uv.lock": "ebeda66f1193fbdddd4a06c7e3ac3c7789d78c84c224259246e43214b7031bfa"
+    }
+  },
   "adapter_weights_changed": false,
   "private_logs_included": false,
   "credentials_included": false