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": 77025.62074989194,
-      "required_pages_per_second_upper_estimate": 0.005328811167541035
+      "required_token_positions_per_second": 76873.35743034528,
+      "required_pages_per_second_upper_estimate": 0.00531827723779522
     },
     {
       "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": 9628.202593736492,
-      "required_pages_per_second_upper_estimate": 0.0040311273662640475
+      "required_token_positions_per_second": 9609.16967879316,
+      "required_pages_per_second_upper_estimate": 0.004023158682229734
     },
     {
       "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": 19256.405187472985,
-      "required_pages_per_second_upper_estimate": 0.004010601374996039
+      "required_token_positions_per_second": 19218.33935758632,
+      "required_pages_per_second_upper_estimate": 0.004002673266494088
     },
     {
       "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": 28884.607781209477,
-      "required_pages_per_second_upper_estimate": 0.0017315881322411447
+      "required_token_positions_per_second": 28827.50903637948,
+      "required_pages_per_second_upper_estimate": 0.0017281651496733215
     },
     {
       "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": 28884.607781209477,
-      "required_pages_per_second_upper_estimate": 0.002245651205026461
+      "required_token_positions_per_second": 28827.50903637948,
+      "required_pages_per_second_upper_estimate": 0.0022412120287667653
     },
     {
       "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": 28884.607781209477,
-      "required_pages_per_second_upper_estimate": 0.3705007347418514
+      "required_token_positions_per_second": 28827.50903637948,
+      "required_pages_per_second_upper_estimate": 0.3697683333510278
     }
   ],
   "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": 0.2129644436563467,
-    "producer_p95_seconds": 0.28998892399977194,
-    "gpu_update_seconds": 0.957188001631307,
-    "producer_batches_per_second": 4.695619526110495,
-    "gpu_batches_per_second": 1.0447268439384214,
-    "producer_headroom": 4.49459067061865,
+    "producer_batch_seconds": 0.08103800960151375,
+    "producer_p95_seconds": 0.30123256300066714,
+    "gpu_update_seconds": 0.9590839071495573,
+    "producer_batches_per_second": 12.33988846613183,
+    "gpu_batches_per_second": 1.0426616404940494,
+    "producer_headroom": 11.834988443887472,
     "sustainable": true,
-    "buffer_seconds": 60.30284410277234,
+    "buffer_seconds": 60.42228615042211,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341384.48172185593,
+    "generation_bytes_per_second": 341632.7762305177,
     "upload_bytes_per_second": 5169525.167320821,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.2531627535034,
+    "estimated_full_shards_per_hour": 73.30644097506188,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -274,3 +274,4 @@
 {"step": 952400, "loss": 0.5081045627593994, "lr": 3e-05, "grad_norm": 0.3509164334995831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139405519742, "tokens_per_second": 152630.40931622678, "data_wait_s": 1.0414998541818932e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952500, "loss": 0.5086105465888977, "lr": 3e-05, "grad_norm": 0.3440548034632097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139420057862, "tokens_per_second": 150938.92247970775, "data_wait_s": 1.4551002095686272e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952600, "loss": 0.5340831279754639, "lr": 3e-05, "grad_norm": 0.3493283224978132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139434654639, "tokens_per_second": 151007.9218448461, "data_wait_s": 1.3346001651370898e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 952700, "loss": 0.5287927985191345, "lr": 3e-05, "grad_norm": 0.3457655496591882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139449234144, "tokens_per_second": 151632.50120178578, "data_wait_s": 9.719999070512131e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9725cd749cb10db232aa57f8ac03b636785e37187760a1ef38e4fae3897a0dfa
-size 588648
+oid sha256:84982f064ed8ac4b64ca43a54803758b60e6eae8c5cc19660fb8df64b1ab69f7
+size 590766