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": 75921.09543785885,
-      "required_pages_per_second_upper_estimate": 0.00658064813229641
+      "required_token_positions_per_second": 75318.70154114274,
+      "required_pages_per_second_upper_estimate": 0.0065284341560824305
     },
     {
       "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": 9490.136929732356,
-      "required_pages_per_second_upper_estimate": 0.0037857575114617664
+      "required_token_positions_per_second": 9414.837692642843,
+      "required_pages_per_second_upper_estimate": 0.0037557195199628382
     },
     {
       "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": 18980.273859464713,
-      "required_pages_per_second_upper_estimate": 0.004390752315976166
+      "required_token_positions_per_second": 18829.675385285685,
+      "required_pages_per_second_upper_estimate": 0.004355914009417469
     },
     {
       "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": 28470.41078919707,
-      "required_pages_per_second_upper_estimate": 0.0017336986443608213
+      "required_token_positions_per_second": 28244.513077928525,
+      "required_pages_per_second_upper_estimate": 0.0017199426589382613
     },
     {
       "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": 28470.41078919707,
-      "required_pages_per_second_upper_estimate": 0.0021917310336935923
+      "required_token_positions_per_second": 28244.513077928525,
+      "required_pages_per_second_upper_estimate": 0.0021743408025552526
     },
     {
       "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": 28470.41078919707,
-      "required_pages_per_second_upper_estimate": 0.3535072176663778
+      "required_token_positions_per_second": 28244.513077928525,
+      "required_pages_per_second_upper_estimate": 0.35070232412342806
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.3213463448591938,
-    "producer_p95_seconds": 0.29545261500243214,
-    "gpu_update_seconds": 0.9711134905890038,
-    "producer_batches_per_second": 3.1119071867401384,
-    "gpu_batches_per_second": 1.0297457606046394,
-    "producer_headroom": 3.0220150505042223,
+    "producer_batch_seconds": 0.16324448264066405,
+    "producer_p95_seconds": 0.2270555940012855,
+    "gpu_update_seconds": 0.9788803908113868,
+    "producer_batches_per_second": 6.125781305584542,
+    "gpu_batches_per_second": 1.0215752704690584,
+    "producer_headroom": 5.996407198435683,
     "sustainable": true,
-    "buffer_seconds": 60.20903641651823,
+    "buffer_seconds": 60.69058423030598,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383846.20888870675,
+    "generation_bytes_per_second": 388842.3072187276,
     "upload_bytes_per_second": 5593351.756579604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.36446094509031,
+    "estimated_full_shards_per_hour": 83.43650734349605,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -188,3 +188,4 @@
 {"step": 333800, "loss": 0.558652114123106, "lr": 0.0003, "grad_norm": 0.15084433480961412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48860349082, "tokens_per_second": 146877.6915620611, "data_wait_s": 1.4217999705579132e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333900, "loss": 0.5932654645293951, "lr": 0.0003, "grad_norm": 0.16861895595353657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48874902633, "tokens_per_second": 147974.48455094354, "data_wait_s": 1.5769997844472528e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334000, "loss": 0.6090955585241318, "lr": 0.0003, "grad_norm": 0.1708305384977711, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48889397255, "tokens_per_second": 146412.8560815567, "data_wait_s": 1.8644997908268124e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 334100, "loss": 0.6207200642675161, "lr": 0.0003, "grad_norm": 0.18417173892827268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48903975511, "tokens_per_second": 147978.44412231215, "data_wait_s": 1.3945998944109306e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e911d80419d7349f38a9ac739f0f58b5200ac92cff555aad2d2355ed576d8438
-size 404934
+oid sha256:62c5affca770e39ffd054d3e27ac6652f2f9946d33ae2132b98b3cb4b8e13c55
+size 407052