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": 77479.28604667087,
-      "required_pages_per_second_upper_estimate": 0.005360196785420716
+      "required_token_positions_per_second": 77545.54883278858,
+      "required_pages_per_second_upper_estimate": 0.0053647809987151865
     },
     {
       "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": 9684.910755833858,
-      "required_pages_per_second_upper_estimate": 0.00405486988953313
+      "required_token_positions_per_second": 9693.193604098573,
+      "required_pages_per_second_upper_estimate": 0.004058337745136026
     },
     {
       "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": 19369.821511667717,
-      "required_pages_per_second_upper_estimate": 0.004034223004336198
+      "required_token_positions_per_second": 19386.387208197146,
+      "required_pages_per_second_upper_estimate": 0.004037673202056482
     },
     {
       "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": 29054.732267501575,
-      "required_pages_per_second_upper_estimate": 0.0017417868354293062
+      "required_token_positions_per_second": 29079.58081229572,
+      "required_pages_per_second_upper_estimate": 0.0017432764677481844
     },
     {
       "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": 29054.732267501575,
-      "required_pages_per_second_upper_estimate": 0.0022588776355371393
+      "required_token_positions_per_second": 29079.58081229572,
+      "required_pages_per_second_upper_estimate": 0.0022608095005976865
     },
     {
       "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": 29054.732267501575,
-      "required_pages_per_second_upper_estimate": 0.37268290898656475
+      "required_token_positions_per_second": 29079.58081229572,
+      "required_pages_per_second_upper_estimate": 0.37300163943889536
     }
   ],
   "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": 0.05079897682812984,
-    "producer_p95_seconds": 0.1458895279997705,
-    "gpu_update_seconds": 0.9515833684320321,
-    "producer_batches_per_second": 19.68543585795712,
-    "gpu_batches_per_second": 1.0508800733326669,
-    "producer_headroom": 18.732333362767545,
+    "producer_batch_seconds": 0.05241596425000594,
+    "producer_p95_seconds": 0.15138905799994973,
+    "gpu_update_seconds": 0.9507702390369257,
+    "producer_batches_per_second": 19.078157090277827,
+    "gpu_batches_per_second": 1.0517788198891682,
+    "producer_headroom": 18.13894397710747,
     "sustainable": true,
-    "buffer_seconds": 60.90133557965005,
+    "buffer_seconds": 60.849295298363245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 687500, "loss": 0.5395483374595642, "lr": 8e-05, "grad_norm": 0.19677089833881958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100631644735, "tokens_per_second": 153012.2355017569, "data_wait_s": 1.9391000023460947e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687600, "loss": 0.5459145903587341, "lr": 8e-05, "grad_norm": 0.2509706096515456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100646213173, "tokens_per_second": 152792.07975177924, "data_wait_s": 1.332200008619111e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687700, "loss": 0.5321168303489685, "lr": 8e-05, "grad_norm": 0.2074190756853795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100660878824, "tokens_per_second": 153923.98432560425, "data_wait_s": 1.5165000149863772e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687800, "loss": 0.5615301728248596, "lr": 8e-05, "grad_norm": 0.20430051949621292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100675435495, "tokens_per_second": 152622.106763418, "data_wait_s": 1.5920999885565834e-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:c2f04d19616dfabeedfe8bc3ab99f1c16beff35c7634c56ffca896b8eadde5dc
-size 57351
+oid sha256:e68674605305bc34545aab77f0d22c18b1d22a076d5c8a04bd6abf6d97b85f6d
+size 59469