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": 76995.87710836295,
-      "required_pages_per_second_upper_estimate": 0.005326753433405303
+      "required_token_positions_per_second": 77042.77186334193,
+      "required_pages_per_second_upper_estimate": 0.005329997721365571
     },
     {
       "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": 9624.48463854537,
-      "required_pages_per_second_upper_estimate": 0.004029570736065258
+      "required_token_positions_per_second": 9630.346482917741,
+      "required_pages_per_second_upper_estimate": 0.0040320249678947396
     },
     {
       "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": 19248.96927709074,
-      "required_pages_per_second_upper_estimate": 0.004009052670961561
+      "required_token_positions_per_second": 19260.692965835482,
+      "required_pages_per_second_upper_estimate": 0.004011494406152628
     },
     {
       "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": 28873.45391563611,
-      "required_pages_per_second_upper_estimate": 0.0017309194750309879
+      "required_token_positions_per_second": 28891.039448753225,
+      "required_pages_per_second_upper_estimate": 0.0017319737008898047
     },
     {
       "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": 28873.45391563611,
-      "required_pages_per_second_upper_estimate": 0.002244784041038802
+      "required_token_positions_per_second": 28891.039448753225,
+      "required_pages_per_second_upper_estimate": 0.0022461512388880717
     },
     {
       "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": 28873.45391563611,
-      "required_pages_per_second_upper_estimate": 0.37035766493036404
+      "required_token_positions_per_second": 28891.039448753225,
+      "required_pages_per_second_upper_estimate": 0.370583233267316
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05167536717187815,
-    "producer_p95_seconds": 0.1416347690001203,
-    "gpu_update_seconds": 0.9575577650246936,
-    "producer_batches_per_second": 19.351579964083975,
-    "gpu_batches_per_second": 1.0443234199810514,
-    "producer_headroom": 18.530255660104892,
+    "producer_batch_seconds": 0.06077427389063317,
+    "producer_p95_seconds": 0.14193187799992302,
+    "gpu_update_seconds": 0.9569749142823981,
+    "producer_batches_per_second": 16.454330689323545,
+    "gpu_batches_per_second": 1.0449594708027063,
+    "producer_headroom": 15.746381700989632,
     "sustainable": true,
-    "buffer_seconds": 60.326139196555694,
+    "buffer_seconds": 60.28941959979108,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 723500, "loss": 0.5432611107826233, "lr": 5e-05, "grad_norm": 0.22921308872240645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105900622546, "tokens_per_second": 152058.41830552442, "data_wait_s": 1.6374000097130192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723600, "loss": 0.5428845882415771, "lr": 5e-05, "grad_norm": 0.21916657527971525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105915173698, "tokens_per_second": 151386.29087988505, "data_wait_s": 1.0311000096407952e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723700, "loss": 0.5550183653831482, "lr": 5e-05, "grad_norm": 0.22080920255256617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105929663900, "tokens_per_second": 150692.5883327505, "data_wait_s": 9.963000138668576e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 723800, "loss": 0.5272859334945679, "lr": 5e-05, "grad_norm": 0.21699936786714963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105944292649, "tokens_per_second": 151765.92871209202, "data_wait_s": 1.1854999684146605e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de31b11f20ef970a7bf55d4faee840993d56a09bd97f4674f5054c108a35d624
-size 78531
+oid sha256:a56d1aef67817063605fb06f3c897367af912205310ca551dec5a82553c72cfe
+size 80649