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": 76793.98366392785,
-      "required_pages_per_second_upper_estimate": 0.005312785976461951
+      "required_token_positions_per_second": 76496.15405442558,
+      "required_pages_per_second_upper_estimate": 0.00529218143301668
     },
     {
       "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": 9599.247957990981,
-      "required_pages_per_second_upper_estimate": 0.004019004664918953
+      "required_token_positions_per_second": 9562.019256803198,
+      "required_pages_per_second_upper_estimate": 0.004003417785155313
     },
     {
       "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": 19198.495915981963,
-      "required_pages_per_second_upper_estimate": 0.003998540400914647
+      "required_token_positions_per_second": 19124.038513606396,
+      "required_pages_per_second_upper_estimate": 0.003983032887573562
     },
     {
       "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": 28797.743873972948,
-      "required_pages_per_second_upper_estimate": 0.0017263807736358214
+      "required_token_positions_per_second": 28686.057770409596,
+      "required_pages_per_second_upper_estimate": 0.0017196853622620043
     },
     {
       "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": 28797.743873972948,
-      "required_pages_per_second_upper_estimate": 0.00223889791831277
+      "required_token_positions_per_second": 28686.057770409596,
+      "required_pages_per_second_upper_estimate": 0.0022302148150159717
     },
     {
       "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": 28797.743873972948,
-      "required_pages_per_second_upper_estimate": 0.36938653780701824
+      "required_token_positions_per_second": 28686.057770409596,
+      "required_pages_per_second_upper_estimate": 0.36795394838970247
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.17096050562497567,
-    "producer_p95_seconds": 0.25604570699942997,
-    "gpu_update_seconds": 0.9600752100926881,
-    "producer_batches_per_second": 5.849304179022677,
-    "gpu_batches_per_second": 1.0415850648861742,
-    "producer_headroom": 5.6157719385712355,
+    "producer_batch_seconds": 0.08136416514845735,
+    "producer_p95_seconds": 0.2163252600003034,
+    "gpu_update_seconds": 0.9638131604308355,
+    "producer_batches_per_second": 12.290422917452622,
+    "gpu_batches_per_second": 1.0375454922746525,
+    "producer_headroom": 11.845671355101581,
     "sustainable": true,
-    "buffer_seconds": 60.48473823583935,
+    "buffer_seconds": 60.720229107142636,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 534534.0209854754,
+    "generation_bytes_per_second": 559760.5575013004,
     "upload_bytes_per_second": 5286647.891867109,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.69855758832165,
+    "estimated_full_shards_per_hour": 60.05579254045133,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 775300, "loss": 0.5524370670318604, "lr": 3e-05, "grad_norm": 0.25227665510208314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113481562555, "tokens_per_second": 151825.27060403733, "data_wait_s": 1.9462000636849552e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775400, "loss": 0.5622367858886719, "lr": 3e-05, "grad_norm": 0.2651238242944428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113496280085, "tokens_per_second": 152814.09469151936, "data_wait_s": 1.0277000910718925e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775500, "loss": 0.5328605771064758, "lr": 3e-05, "grad_norm": 0.24942127565431396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113511064431, "tokens_per_second": 153433.6788007891, "data_wait_s": 1.2311000318732113e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 775600, "loss": 0.5252447128295898, "lr": 3e-05, "grad_norm": 0.2516212729134852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113525843560, "tokens_per_second": 153074.31958093596, "data_wait_s": 1.1676998838083819e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0432caef8b873ba8464ff4240560d00e2813d5db57472e4446b97373898684da
-size 330804
+oid sha256:66052ab4a6034fb24663b511c130f351646aa6700d68b1ad096d7a6a18a69bbe
+size 332922