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": 77498.18788218389,
-      "required_pages_per_second_upper_estimate": 0.005361504458259808
+      "required_token_positions_per_second": 77453.15198894475,
+      "required_pages_per_second_upper_estimate": 0.0053583887706678554
     },
     {
       "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": 9687.273485272986,
-      "required_pages_per_second_upper_estimate": 0.004055859115009892
+      "required_token_positions_per_second": 9681.643998618094,
+      "required_pages_per_second_upper_estimate": 0.004053502166504537
     },
     {
       "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": 19374.546970545973,
-      "required_pages_per_second_upper_estimate": 0.004035207192801808
+      "required_token_positions_per_second": 19363.287997236188,
+      "required_pages_per_second_upper_estimate": 0.0040328622455804725
     },
     {
       "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": 29061.82045581896,
-      "required_pages_per_second_upper_estimate": 0.0017422117614958967
+      "required_token_positions_per_second": 29044.93199585428,
+      "required_pages_per_second_upper_estimate": 0.0017411993241081994
     },
     {
       "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": 29061.82045581896,
-      "required_pages_per_second_upper_estimate": 0.0022594287110011707
+      "required_token_positions_per_second": 29044.93199585428,
+      "required_pages_per_second_upper_estimate": 0.0022581157075234016
     },
     {
       "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": 29061.82045581896,
-      "required_pages_per_second_upper_estimate": 0.37277382865559655
+      "required_token_positions_per_second": 29044.93199585428,
+      "required_pages_per_second_upper_estimate": 0.3725572016245851
     }
   ],
   "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": 0.046973330882810416,
-    "producer_p95_seconds": 0.12575947599998472,
-    "gpu_update_seconds": 0.9513512769109453,
-    "producer_batches_per_second": 21.2886755358016,
-    "gpu_batches_per_second": 1.0511364458846557,
-    "producer_headroom": 20.253008654727658,
+    "producer_batch_seconds": 0.05182851983593473,
+    "producer_p95_seconds": 0.1311697619998995,
+    "gpu_update_seconds": 0.951904449421549,
+    "producer_batches_per_second": 19.294396273818748,
+    "gpu_batches_per_second": 1.0505256074889424,
+    "producer_headroom": 18.366421661950618,
     "sustainable": true,
-    "buffer_seconds": 60.8864817223005,
+    "buffer_seconds": 60.92188476297913,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 72,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -12,3 +12,4 @@
 {"step": 686200, "loss": 0.5173473358154297, "lr": 8e-05, "grad_norm": 0.21569271361113784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100441963613, "tokens_per_second": 155056.08306677538, "data_wait_s": 1.6467000023112632e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686300, "loss": 0.5384304523468018, "lr": 8e-05, "grad_norm": 0.19933378543151312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100456651841, "tokens_per_second": 154290.50755349858, "data_wait_s": 1.608100001249113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686400, "loss": 0.5408609509468079, "lr": 8e-05, "grad_norm": 0.19459340893772095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100471198826, "tokens_per_second": 152717.55133803163, "data_wait_s": 1.5714000028310693e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 686500, "loss": 0.5647568702697754, "lr": 8e-05, "grad_norm": 0.21216946503323078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100485727691, "tokens_per_second": 151653.19817562008, "data_wait_s": 9.998999985327828e-06, "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 sha256:537edb1eef747e851677a1e91f7fda386e3a5407d988fcc798421141ce37fa0a
-size 29817
+oid sha256:f995e78f1337b05e928e7c21ecf0cf8673c14df400e89d8ead22b9f81a1f0809
+size 31935