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": 76411.80871529346,
-      "required_pages_per_second_upper_estimate": 0.00528634622674737
+      "required_token_positions_per_second": 76357.51005666674,
+      "required_pages_per_second_upper_estimate": 0.005282589719553327
     },
     {
       "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": 9551.476089411683,
-      "required_pages_per_second_upper_estimate": 0.003999003581134856
+      "required_token_positions_per_second": 9544.688757083342,
+      "required_pages_per_second_upper_estimate": 0.003996161866824596
     },
     {
       "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": 19102.952178823365,
-      "required_pages_per_second_upper_estimate": 0.0039786411601223
+      "required_token_positions_per_second": 19089.377514166685,
+      "required_pages_per_second_upper_estimate": 0.0039758139154622936
     },
     {
       "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": 28654.42826823505,
-      "required_pages_per_second_upper_estimate": 0.0017177892219020937
+      "required_token_positions_per_second": 28634.066271250027,
+      "required_pages_per_second_upper_estimate": 0.0017165685512738108
     },
     {
       "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": 28654.42826823505,
-      "required_pages_per_second_upper_estimate": 0.002227755760345378
+      "required_token_positions_per_second": 28634.066271250027,
+      "required_pages_per_second_upper_estimate": 0.002226172704642752
     },
     {
       "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": 28654.42826823505,
-      "required_pages_per_second_upper_estimate": 0.3675482390969209
+      "required_token_positions_per_second": 28634.066271250027,
+      "required_pages_per_second_upper_estimate": 0.367287057262606
     }
   ],
   "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": 0.12851214425762691,
-    "producer_p95_seconds": 0.2908599010006583,
-    "gpu_update_seconds": 0.964877042430795,
-    "producer_batches_per_second": 7.781365767232946,
-    "gpu_batches_per_second": 1.036401485396233,
-    "producer_headroom": 7.508061187559958,
+    "producer_batch_seconds": 0.08469342544537994,
+    "producer_p95_seconds": 0.2953102920000674,
+    "gpu_update_seconds": 0.965563177024561,
+    "producer_batches_per_second": 11.80729194434242,
+    "gpu_batches_per_second": 1.0356650127043556,
+    "producer_headroom": 11.400686321835773,
     "sustainable": true,
-    "buffer_seconds": 60.787253673140086,
+    "buffer_seconds": 60.83048015254734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332014.9381710065,
+    "generation_bytes_per_second": 331883.15068931517,
     "upload_bytes_per_second": 5613899.857854061,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.24267681930205,
+    "estimated_full_shards_per_hour": 71.21439829358665,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 674700, "loss": 0.5378531217575073, "lr": 0.0001, "grad_norm": 0.19060410120226906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98758370154, "tokens_per_second": 153270.14406059575, "data_wait_s": 1.1246000212850049e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 674800, "loss": 0.5367597937583923, "lr": 0.0001, "grad_norm": 0.1808463882216845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98773094831, "tokens_per_second": 152515.78546607023, "data_wait_s": 1.2541997421067208e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 674900, "loss": 0.5363932251930237, "lr": 0.0001, "grad_norm": 0.20466907236173054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98787737477, "tokens_per_second": 152134.1510004654, "data_wait_s": 8.256000000983477e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 675000, "loss": 0.5931280255317688, "lr": 0.0001, "grad_norm": 0.20633527339194027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98802239020, "tokens_per_second": 151837.96994796887, "data_wait_s": 1.0897001629928127e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9debbe04d6bd5c13f00a0247fbab803ad4e51ff0c366de6e867c4352881b07b
-size 637362
+oid sha256:0d423cbeee6dcf9f9f2b5789094ed39b34d3bd47940a2535e02fb3758557a89f
+size 639480