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": 77558.95367787528,
-      "required_pages_per_second_upper_estimate": 0.00536570837700181
+      "required_token_positions_per_second": 77681.33663039813,
+      "required_pages_per_second_upper_estimate": 0.005374175113624913
     },
     {
       "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": 9694.86920973441,
-      "required_pages_per_second_upper_estimate": 0.0040590392862251265
+      "required_token_positions_per_second": 9710.167078799766,
+      "required_pages_per_second_upper_estimate": 0.004065444184546958
     },
     {
       "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": 19389.73841946882,
-      "required_pages_per_second_upper_estimate": 0.004038371170986988
+      "required_token_positions_per_second": 19420.334157599533,
+      "required_pages_per_second_upper_estimate": 0.004044743456375742
     },
     {
       "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": 29084.607629203234,
-      "required_pages_per_second_upper_estimate": 0.0017435778177462848
+      "required_token_positions_per_second": 29130.5012363993,
+      "required_pages_per_second_upper_estimate": 0.0017463290694222082
     },
     {
       "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": 29084.607629203234,
-      "required_pages_per_second_upper_estimate": 0.002261200313501615
+      "required_token_positions_per_second": 29130.5012363993,
+      "required_pages_per_second_upper_estimate": 0.0022647683395964646
     },
     {
       "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": 29084.607629203234,
-      "required_pages_per_second_upper_estimate": 0.37306611804880946
+      "required_token_positions_per_second": 29130.5012363993,
+      "required_pages_per_second_upper_estimate": 0.3736547919652044
     }
   ],
   "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": 0.05344180156249845,
-    "producer_p95_seconds": 0.14228422400003637,
-    "gpu_update_seconds": 0.9506059133573882,
-    "producer_batches_per_second": 18.711944035616625,
-    "gpu_batches_per_second": 1.0519606347368067,
-    "producer_headroom": 17.78768465066967,
+    "producer_batch_seconds": 0.0466080403906286,
+    "producer_p95_seconds": 0.12629458900005375,
+    "gpu_update_seconds": 0.9491082826083721,
+    "producer_batches_per_second": 21.455525519177765,
+    "gpu_batches_per_second": 1.0536205597655997,
+    "producer_headroom": 20.36361697796691,
     "sustainable": true,
-    "buffer_seconds": 60.838778454872845,
+    "buffer_seconds": 60.742930086935814,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 51,
+    "upload_pages_per_commit": 58,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -3,3 +3,4 @@
 {"step": 685300, "loss": 0.6020703315734863, "lr": 8e-05, "grad_norm": 0.2038634766871492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100310140405, "tokens_per_second": 152199.12537603173, "data_wait_s": 1.5465999922525953e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685400, "loss": 0.5669589638710022, "lr": 8e-05, "grad_norm": 0.1897535348704872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100324882735, "tokens_per_second": 154863.5560354908, "data_wait_s": 1.8494000073587813e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685500, "loss": 0.5513772368431091, "lr": 8e-05, "grad_norm": 0.1995322552956654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100339543873, "tokens_per_second": 154138.54795256618, "data_wait_s": 1.2913000091430149e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685600, "loss": 0.5076792240142822, "lr": 8e-05, "grad_norm": 0.19537428987943656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100354050683, "tokens_per_second": 151189.80668697378, "data_wait_s": 2.0197000139887678e-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 sha256:4e6f1dc52d1424c4d6c4d09d2777c31f49278c7522a81b17f259470bb1611a26
-size 10755
+oid sha256:dd5a82910a9ceb3237a8d88b8856865aa878b78c02035ae052b968c6e7cdd8f3
+size 12873