Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75697.88887468546,
-      "required_pages_per_second_upper_estimate": 0.006561301153112402
+      "required_token_positions_per_second": 76181.04485715211,
+      "required_pages_per_second_upper_estimate": 0.00660317988912497
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9462.236109335683,
-      "required_pages_per_second_upper_estimate": 0.003774627457051094
+      "required_token_positions_per_second": 9522.630607144014,
+      "required_pages_per_second_upper_estimate": 0.003798719725205048
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18924.472218671366,
-      "required_pages_per_second_upper_estimate": 0.004377843588453961
+      "required_token_positions_per_second": 19045.261214288028,
+      "required_pages_per_second_upper_estimate": 0.004405785996776158
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28386.708328007047,
-      "required_pages_per_second_upper_estimate": 0.0017286016036272233
+      "required_token_positions_per_second": 28567.89182143204,
+      "required_pages_per_second_upper_estimate": 0.0017396347277804812
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28386.708328007047,
-      "required_pages_per_second_upper_estimate": 0.002185287386527942
+      "required_token_positions_per_second": 28567.89182143204,
+      "required_pages_per_second_upper_estimate": 0.0021992353933998083
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28386.708328007047,
-      "required_pages_per_second_upper_estimate": 0.35246791323251486
+      "required_token_positions_per_second": 28567.89182143204,
+      "required_pages_per_second_upper_estimate": 0.35471760583870815
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13977126414047802,
-    "producer_p95_seconds": 0.32535314800043125,
-    "gpu_update_seconds": 0.9739769641667745,
-    "producer_batches_per_second": 7.15454643806429,
-    "gpu_batches_per_second": 1.026718327835903,
-    "producer_headroom": 6.968363419736066,
+    "producer_batch_seconds": 0.2837696983907847,
+    "producer_p95_seconds": 0.3436622160006664,
+    "gpu_update_seconds": 0.9677998003079133,
+    "producer_batches_per_second": 3.523984434105719,
+    "gpu_batches_per_second": 1.0332715502543417,
+    "producer_headroom": 3.41051143161571,
     "sustainable": true,
-    "buffer_seconds": 60.38657177834002,
+    "buffer_seconds": 60.00358761909063,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334235.1722691521,
+    "generation_bytes_per_second": 326061.4282588636,
     "upload_bytes_per_second": 4468486.906889906,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.7190873723595,
+    "estimated_full_shards_per_hour": 69.96519218277389,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -280,3 +280,4 @@
 {"step": 183000, "loss": 0.6215140540152788, "lr": 0.0003, "grad_norm": 0.13817083573035038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26783605310, "tokens_per_second": 148652.63892421804, "data_wait_s": 1.3806999049847946e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 183100, "loss": 0.590689055621624, "lr": 0.0003, "grad_norm": 0.1417717697950723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26798195675, "tokens_per_second": 148874.71129757358, "data_wait_s": 1.8180999177275226e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 183200, "loss": 0.5766947977244854, "lr": 0.0003, "grad_norm": 0.15148524851117368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26812852665, "tokens_per_second": 149902.66337086423, "data_wait_s": 1.765200067893602e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 183300, "loss": 0.6264513321220875, "lr": 0.0003, "grad_norm": 0.13549497794452597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26827469125, "tokens_per_second": 149075.66611052203, "data_wait_s": 1.292799788643606e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff730b42a6be389dc4b82a3e0e25d01f7ba78db472b149181d3afc8123e7868b
-size 601356
+oid sha256:149f9b9deb27f67bbfde0433c551f6db2233cf87719299b67d85665b654587c6
+size 603474