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": 76245.46945138989,
-      "required_pages_per_second_upper_estimate": 0.0052748384905102535
+      "required_token_positions_per_second": 77049.96630814987,
+      "required_pages_per_second_upper_estimate": 0.005330495449750797
     },
     {
       "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": 9530.683681423736,
-      "required_pages_per_second_upper_estimate": 0.003990298234105155
+      "required_token_positions_per_second": 9631.245788518734,
+      "required_pages_per_second_upper_estimate": 0.004032401488370239
     },
     {
       "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": 19061.36736284747,
-      "required_pages_per_second_upper_estimate": 0.003969980139619866
+      "required_token_positions_per_second": 19262.49157703747,
+      "required_pages_per_second_upper_estimate": 0.004011869009433435
     },
     {
       "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": 28592.051044271208,
-      "required_pages_per_second_upper_estimate": 0.0017140498025699676
+      "required_token_positions_per_second": 28893.7373655562,
+      "required_pages_per_second_upper_estimate": 0.0017321354368826627
     },
     {
       "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": 28592.051044271208,
-      "required_pages_per_second_upper_estimate": 0.002222906205550601
+      "required_token_positions_per_second": 28893.7373655562,
+      "required_pages_per_second_upper_estimate": 0.0022463609900526212
     },
     {
       "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": 28592.051044271208,
-      "required_pages_per_second_upper_estimate": 0.3667481310433577
+      "required_token_positions_per_second": 28893.7373655562,
+      "required_pages_per_second_upper_estimate": 0.3706178392472672
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11466135003908562,
-    "producer_p95_seconds": 0.20930288700037636,
-    "gpu_update_seconds": 0.9669820453660543,
-    "producer_batches_per_second": 8.721334605419536,
-    "gpu_batches_per_second": 1.0341453647378185,
-    "producer_headroom": 8.433373975070332,
+    "producer_batch_seconds": 0.06667956513285844,
+    "producer_p95_seconds": 0.20637315000021772,
+    "gpu_update_seconds": 0.9568855579395822,
+    "producer_batches_per_second": 14.997098406498436,
+    "gpu_batches_per_second": 1.0450570517055917,
+    "producer_headroom": 14.350506876177075,
     "sustainable": true,
-    "buffer_seconds": 60.91986885806142,
+    "buffer_seconds": 60.28379015019368,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385092.3134157955,
-    "upload_bytes_per_second": 3586563.771010249,
+    "generation_bytes_per_second": 393828.15701383096,
+    "upload_bytes_per_second": 3615412.57026016,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.63184596877478,
+    "estimated_full_shards_per_hour": 84.50635464488217,
     "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": [
-    {
-      "seconds": 2.0315385989997594,
-      "bytes": 5979371,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.26191803599977,
       "bytes": 2574745,
@@ -562,6 +556,12 @@
       "bytes": 1195587,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4843279429987888,
+      "bytes": 7415076,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -160,3 +160,4 @@
 {"step": 941000, "loss": 0.5244556069374084, "lr": 3e-05, "grad_norm": 0.336522169660098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137735627709, "tokens_per_second": 153399.61329272506, "data_wait_s": 1.2283999240025878e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941100, "loss": 0.5392598509788513, "lr": 3e-05, "grad_norm": 0.36996917852931727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137750354229, "tokens_per_second": 153107.729645144, "data_wait_s": 1.302700002270285e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941200, "loss": 0.5398966670036316, "lr": 3e-05, "grad_norm": 0.3430767914103548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137765008400, "tokens_per_second": 153301.6113073163, "data_wait_s": 1.3201999536249787e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 941300, "loss": 0.500248908996582, "lr": 3e-05, "grad_norm": 0.3266684065896893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137779724167, "tokens_per_second": 153474.19305677293, "data_wait_s": 1.4176999684423208e-05, "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 sha256:533dfa381b64ee8d26c8f331b60c590c60b137d3ec26bddaba2eea30839ad27e
-size 345630
+oid sha256:61d6cd5c2b6ec5f22c45e4238b66f7d83b54f040e9e0cc3654dc667c568ec598
+size 347748