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": 77723.0996774248,
-      "required_pages_per_second_upper_estimate": 0.0053770643781218385
+      "required_token_positions_per_second": 77603.70305075262,
+      "required_pages_per_second_upper_estimate": 0.005368804242450315
     },
     {
       "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": 9715.3874596781,
-      "required_pages_per_second_upper_estimate": 0.004067629849006768
+      "required_token_positions_per_second": 9700.462881344078,
+      "required_pages_per_second_upper_estimate": 0.004061381239718948
     },
     {
       "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": 19430.7749193562,
-      "required_pages_per_second_upper_estimate": 0.0040469179917082525
+      "required_token_positions_per_second": 19400.925762688155,
+      "required_pages_per_second_upper_estimate": 0.004040701199549495
     },
     {
       "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": 29146.1623790343,
-      "required_pages_per_second_upper_estimate": 0.0017472679310099964
+      "required_token_positions_per_second": 29101.38864403223,
+      "required_pages_per_second_upper_estimate": 0.001744583813962157
     },
     {
       "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": 29146.1623790343,
-      "required_pages_per_second_upper_estimate": 0.0022659859245502496
+      "required_token_positions_per_second": 29101.38864403223,
+      "required_pages_per_second_upper_estimate": 0.002262504963592686
     },
     {
       "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": 29146.1623790343,
-      "required_pages_per_second_upper_estimate": 0.3738556762873013
+      "required_token_positions_per_second": 29101.38864403223,
+      "required_pages_per_second_upper_estimate": 0.37328136688898594
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.049101370765625596,
-    "producer_p95_seconds": 0.11197897300007753,
-    "gpu_update_seconds": 0.9485982971085081,
-    "producer_batches_per_second": 20.36603020256351,
-    "gpu_batches_per_second": 1.0541870073435438,
-    "producer_headroom": 19.31918156901219,
+    "producer_batch_seconds": 0.04587113568749768,
+    "producer_p95_seconds": 0.12252099700003782,
+    "gpu_update_seconds": 0.9500577562875071,
+    "producer_batches_per_second": 21.800201477735662,
+    "gpu_batches_per_second": 1.0525675869513973,
+    "producer_headroom": 20.71145050255314,
     "sustainable": true,
-    "buffer_seconds": 60.71029101494452,
+    "buffer_seconds": 60.803696402400455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 42,
+    "upload_pages_per_commit": 48,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 645100, "loss": 0.5672259330749512, "lr": 0.0001, "grad_norm": 0.19234217093846068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94425964227, "tokens_per_second": 153596.82575021993, "data_wait_s": 1.702800000202842e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 645200, "loss": 0.5683538913726807, "lr": 0.0001, "grad_norm": 0.18104020726054085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94440612732, "tokens_per_second": 154254.8723846061, "data_wait_s": 1.5940000025693735e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 645300, "loss": 0.5487010478973389, "lr": 0.0001, "grad_norm": 0.17951284983949858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94455290557, "tokens_per_second": 154418.1227540069, "data_wait_s": 1.4860999954180443e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 645400, "loss": 0.516700029373169, "lr": 0.0001, "grad_norm": 0.16708887996363114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94470019608, "tokens_per_second": 154878.31277059382, "data_wait_s": 9.39000005928392e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b99803909dc1a1b9710991f3f437f302713c4ce53a004c63aeed3ebe1a5cad1
-size 6519
+oid sha256:3757d97292ca9e78d4fc5d4532daa5fa1fb049fcba499674e518dc1acf6d993b
+size 8637