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": 77015.12550654053,
-      "required_pages_per_second_upper_estimate": 0.005328085082253665
+      "required_token_positions_per_second": 77188.07918325684,
+      "required_pages_per_second_upper_estimate": 0.005340050418916718
     },
     {
       "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": 9626.890688317566,
-      "required_pages_per_second_upper_estimate": 0.004030578098860844
+      "required_token_positions_per_second": 9648.509897907104,
+      "required_pages_per_second_upper_estimate": 0.00403962961045555
     },
     {
       "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": 19253.781376635132,
-      "required_pages_per_second_upper_estimate": 0.0040100549043930594
+      "required_token_positions_per_second": 19297.01979581421,
+      "required_pages_per_second_upper_estimate": 0.004019060326834268
     },
     {
       "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": 28880.6720649527,
-      "required_pages_per_second_upper_estimate": 0.0017313521920610418
+      "required_token_positions_per_second": 28945.52969372131,
+      "required_pages_per_second_upper_estimate": 0.0017352403078738548
     },
     {
       "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": 28880.6720649527,
-      "required_pages_per_second_upper_estimate": 0.0022453452203988833
+      "required_token_positions_per_second": 28945.52969372131,
+      "required_pages_per_second_upper_estimate": 0.002250387615757083
     },
     {
       "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": 28880.6720649527,
-      "required_pages_per_second_upper_estimate": 0.37045025159955236
+      "required_token_positions_per_second": 28945.52969372131,
+      "required_pages_per_second_upper_estimate": 0.37128217562269994
     }
   ],
   "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": 0.08503467300791101,
-    "producer_p95_seconds": 0.19995305299926258,
-    "gpu_update_seconds": 0.9573184425147581,
-    "producer_batches_per_second": 11.759908806928289,
-    "gpu_batches_per_second": 1.0445844930900137,
-    "producer_headroom": 11.257977583164177,
+    "producer_batch_seconds": 0.3809135796095262,
+    "producer_p95_seconds": 0.2068906380009139,
+    "gpu_update_seconds": 0.9551733995732418,
+    "producer_batches_per_second": 2.6252673927380012,
+    "gpu_batches_per_second": 1.0469303274638784,
+    "producer_headroom": 2.5075855803103377,
     "sustainable": true,
-    "buffer_seconds": 60.311061878429754,
+    "buffer_seconds": 60.17592417311423,
     "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": 313106.21688319644,
+    "generation_bytes_per_second": 297539.0138902272,
     "upload_bytes_per_second": 3212749.5799009884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.18530540344162,
+    "estimated_full_shards_per_hour": 63.84494602708923,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 696300, "loss": 0.5331386923789978, "lr": 8e-05, "grad_norm": 0.20653918218985734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101919565504, "tokens_per_second": 134281.24536770766, "data_wait_s": 1.8845999875338748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696400, "loss": 0.5500418543815613, "lr": 8e-05, "grad_norm": 0.21281361190886006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101934368019, "tokens_per_second": 154804.55478760018, "data_wait_s": 1.4496999938273802e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696500, "loss": 0.5282445549964905, "lr": 8e-05, "grad_norm": 0.2143935001575543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101949046168, "tokens_per_second": 153406.3899245804, "data_wait_s": 1.0687999747460708e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 696600, "loss": 0.5328052043914795, "lr": 8e-05, "grad_norm": 0.20013354791740276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101963504039, "tokens_per_second": 150144.31962397718, "data_wait_s": 1.7787000615498982e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:578faee35374839978d0f3871a274bfc0ef4d80ec9981537703a10f47e0d5d4b
-size 245301
+oid sha256:7f44500c25de9dfe36a7cb4ee79687384a0652c6f70f47f5ff3033fdbe17df71
+size 247419