Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76789.04715621483,
-      "required_pages_per_second_upper_estimate": 0.005312444457404088
+      "required_token_positions_per_second": 77167.02178076521,
+      "required_pages_per_second_upper_estimate": 0.005338593618952442
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9598.630894526854,
-      "required_pages_per_second_upper_estimate": 0.004018746313332273
+      "required_token_positions_per_second": 9645.877722595651,
+      "required_pages_per_second_upper_estimate": 0.004038527573618715
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19197.261789053708,
-      "required_pages_per_second_upper_estimate": 0.003998283364821607
+      "required_token_positions_per_second": 19291.755445191302,
+      "required_pages_per_second_upper_estimate": 0.004017963901429778
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28795.89268358056,
-      "required_pages_per_second_upper_estimate": 0.001726269797598402
+      "required_token_positions_per_second": 28937.633167786953,
+      "required_pages_per_second_upper_estimate": 0.001734766923201387
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28795.89268358056,
-      "required_pages_per_second_upper_estimate": 0.0022387539964022884
+      "required_token_positions_per_second": 28937.633167786953,
+      "required_pages_per_second_upper_estimate": 0.0022497736955988334
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28795.89268358056,
-      "required_pages_per_second_upper_estimate": 0.3693627927243181
+      "required_token_positions_per_second": 28937.633167786953,
+      "required_pages_per_second_upper_estimate": 0.37118088746664296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1010391160392885,
+    "producer_batch_seconds": 0.20629423607053354,
     "producer_p95_seconds": 0.2667576369967719,
-    "gpu_update_seconds": 0.9601369300756184,
-    "producer_batches_per_second": 9.897157053622237,
-    "gpu_batches_per_second": 1.0415181092151533,
-    "producer_headroom": 9.502625989941109,
+    "gpu_update_seconds": 0.9554340481023667,
+    "producer_batches_per_second": 4.847445178536605,
+    "gpu_batches_per_second": 1.0466447181635905,
+    "producer_headroom": 4.631414169883528,
     "sustainable": true,
-    "buffer_seconds": 60.48862659476396,
+    "buffer_seconds": 60.1923450304491,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493280.68129095336,
+    "generation_bytes_per_second": 482122.99305443995,
     "upload_bytes_per_second": 5574188.752891006,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.84655121847582,
+    "estimated_full_shards_per_hour": 103.45237106060885,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 505900, "loss": 0.5876098871231079, "lr": 0.0003, "grad_norm": 0.1824659212107423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74046714407, "tokens_per_second": 151443.90136277847, "data_wait_s": 1.7674999980954453e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506000, "loss": 0.5967216491699219, "lr": 0.0003, "grad_norm": 0.15872306353266838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74061387830, "tokens_per_second": 153078.95207051752, "data_wait_s": 1.205899752676487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506100, "loss": 0.5581861138343811, "lr": 0.0003, "grad_norm": 0.1638250157456863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74076017377, "tokens_per_second": 152253.9811607355, "data_wait_s": 1.5487999917240813e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 506200, "loss": 0.6603443622589111, "lr": 0.0003, "grad_norm": 0.17462372872940857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74090805709, "tokens_per_second": 154048.85909780825, "data_wait_s": 1.032499858411029e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a49009a6020f67d5bd0eae9ca40ea30bf63970e69fffc7cac2960cc145752b6
-size 450195
+oid sha256:f5fc49a33ee77ae5e2c45077787f098a1cb37d885c15b489592609b58c9f14fb
+size 452313