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": 75673.68981088143,
-      "required_pages_per_second_upper_estimate": 0.0052352814482683275
+      "required_token_positions_per_second": 76574.21076389805,
+      "required_pages_per_second_upper_estimate": 0.005297581577294505
     },
     {
       "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": 9459.21122636018,
-      "required_pages_per_second_upper_estimate": 0.003960374209684626
+      "required_token_positions_per_second": 9571.776345487257,
+      "required_pages_per_second_upper_estimate": 0.004007502874436146
     },
     {
       "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": 18918.42245272036,
-      "required_pages_per_second_upper_estimate": 0.00394020848455117
+      "required_token_positions_per_second": 19143.552690974513,
+      "required_pages_per_second_upper_estimate": 0.003987097176095876
     },
     {
       "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": 28377.63367908054,
-      "required_pages_per_second_upper_estimate": 0.0017011958089231467
+      "required_token_positions_per_second": 28715.32903646177,
+      "required_pages_per_second_upper_estimate": 0.0017214401299671968
     },
     {
       "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": 28377.63367908054,
-      "required_pages_per_second_upper_estimate": 0.002206236198529344
+      "required_token_positions_per_second": 28715.32903646177,
+      "required_pages_per_second_upper_estimate": 0.0022324905272008357
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28377.63367908054,
-      "required_pages_per_second_upper_estimate": 0.3639978153061215
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28715.32903646177,
+      "required_pages_per_second_upper_estimate": 0.368329408761583
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09526632713283334,
-    "producer_p95_seconds": 0.2706608199987386,
-    "gpu_update_seconds": 0.9742884242099998,
-    "producer_batches_per_second": 10.496888356004984,
-    "gpu_batches_per_second": 1.0263901070269292,
-    "producer_headroom": 10.226996815480392,
+    "producer_batch_seconds": 0.10729765024194649,
+    "producer_p95_seconds": 0.283750554001017,
+    "gpu_update_seconds": 0.9628306875708611,
+    "producer_batches_per_second": 9.319868587476897,
+    "gpu_batches_per_second": 1.0386042041544332,
+    "producer_headroom": 8.973455480150452,
     "sustainable": true,
-    "buffer_seconds": 60.40588230101999,
+    "buffer_seconds": 60.658333316964246,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354665.10162553855,
+    "generation_bytes_per_second": 371434.4236794177,
     "upload_bytes_per_second": 3241048.3318986474,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.10287462782495,
+    "estimated_full_shards_per_hour": 79.70118077074908,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 828300, "loss": 0.5455509424209595, "lr": 3e-05, "grad_norm": 0.2858742443707635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121239761221, "tokens_per_second": 150085.2846416131, "data_wait_s": 1.0996001947205514e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 828400, "loss": 0.5238282084465027, "lr": 3e-05, "grad_norm": 0.29655192923119195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121254383670, "tokens_per_second": 151297.49738241534, "data_wait_s": 1.3859997125109658e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 828500, "loss": 0.518890917301178, "lr": 3e-05, "grad_norm": 0.27430080897680775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121269045734, "tokens_per_second": 152040.20547331744, "data_wait_s": 9.425999451195821e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 828600, "loss": 0.48698189854621887, "lr": 3e-05, "grad_norm": 0.2820801254666777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121283689201, "tokens_per_second": 151651.99115701005, "data_wait_s": 1.44199984788429e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c9d1eba948f01a02da22b235879458e9ace520bc6667ab2fce55a02c4de8ab92
-size 607710
+oid sha256:3e21f66ea92a57e9d75b66d81a6657ce39663a7239a40db47c85623c00c26701
+size 609828