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": 76794.4591712007,
-      "required_pages_per_second_upper_estimate": 0.005312818873158415
+      "required_token_positions_per_second": 77065.1346587177,
+      "required_pages_per_second_upper_estimate": 0.005331544831438501
     },
     {
       "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": 9599.307396400087,
-      "required_pages_per_second_upper_estimate": 0.00401902955053963
+      "required_token_positions_per_second": 9633.141832339712,
+      "required_pages_per_second_upper_estimate": 0.0040331953223241845
     },
     {
       "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": 19198.614792800174,
-      "required_pages_per_second_upper_estimate": 0.003998565159820887
+      "required_token_positions_per_second": 19266.283664679424,
+      "required_pages_per_second_upper_estimate": 0.0040126588012851785
     },
     {
       "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": 28797.92218920026,
-      "required_pages_per_second_upper_estimate": 0.0017263914633614272
+      "required_token_positions_per_second": 28899.42549701914,
+      "required_pages_per_second_upper_estimate": 0.001732476431678592
     },
     {
       "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": 28797.92218920026,
-      "required_pages_per_second_upper_estimate": 0.002238911781537368
+      "required_token_positions_per_second": 28899.42549701914,
+      "required_pages_per_second_upper_estimate": 0.002246803217254419
     },
     {
       "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": 28797.92218920026,
-      "required_pages_per_second_upper_estimate": 0.36938882504329423
+      "required_token_positions_per_second": 28899.42549701914,
+      "required_pages_per_second_upper_estimate": 0.37069080049023406
     }
   ],
   "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": 0.4044776608901657,
-    "producer_p95_seconds": 0.3602808139985427,
-    "gpu_update_seconds": 0.9600692653572243,
-    "producer_batches_per_second": 2.472324424046613,
-    "gpu_batches_per_second": 1.0415915143663288,
-    "producer_headroom": 2.373602693519154,
+    "producer_batch_seconds": 0.2667638340626013,
+    "producer_p95_seconds": 0.363993165003194,
+    "gpu_update_seconds": 0.9566972188721116,
+    "producer_batches_per_second": 3.748634081205066,
+    "gpu_batches_per_second": 1.045262785627139,
+    "producer_headroom": 3.5863078000580995,
     "sustainable": true,
-    "buffer_seconds": 60.484363717505126,
+    "buffer_seconds": 60.27192478894303,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415109.2820839442,
+    "generation_bytes_per_second": 392197.39960015623,
     "upload_bytes_per_second": 5022611.3694364345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.07278868569131,
+    "estimated_full_shards_per_hour": 84.15643206599727,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -384,3 +384,4 @@
 {"step": 478600, "loss": 0.5615736842155457, "lr": 0.0003, "grad_norm": 0.1617892725936926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70051902110, "tokens_per_second": 152980.99716700756, "data_wait_s": 1.5670004358980805e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 478700, "loss": 0.565996527671814, "lr": 0.0003, "grad_norm": 0.15105829437840865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70066633918, "tokens_per_second": 153141.2903115773, "data_wait_s": 1.0431998816784471e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 478800, "loss": 0.5429500341415405, "lr": 0.0003, "grad_norm": 0.17888349647095725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70081062930, "tokens_per_second": 149682.05512863398, "data_wait_s": 9.66299558058381e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 478900, "loss": 0.55815589427948, "lr": 0.0003, "grad_norm": 0.1470691740053245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70095597952, "tokens_per_second": 150136.12578148805, "data_wait_s": 1.3281998690217733e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fcf2616163b7d87e654c82c53189c6a678c0aa9f0ae6e68311bf30ad3f4dac5c
-size 823194
+oid sha256:dab12756c09520ee1ac68ff61f698cb3cb17379f6bd7605e040df65764b87df2
+size 825312