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": 77506.75420997858,
-      "required_pages_per_second_upper_estimate": 0.005362097096693263
+      "required_token_positions_per_second": 77568.99825122934,
+      "required_pages_per_second_upper_estimate": 0.005366403283893065
     },
     {
       "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": 9688.344276247322,
-      "required_pages_per_second_upper_estimate": 0.004056307432830188
+      "required_token_positions_per_second": 9696.124781403667,
+      "required_pages_per_second_upper_estimate": 0.0040595649678637265
     },
     {
       "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": 19376.688552494645,
-      "required_pages_per_second_upper_estimate": 0.004035653227844402
+      "required_token_positions_per_second": 19392.249562807334,
+      "required_pages_per_second_upper_estimate": 0.0040388941759210975
     },
     {
       "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": 29065.032828741965,
-      "required_pages_per_second_upper_estimate": 0.0017424043383476213
+      "required_token_positions_per_second": 29088.374344211003,
+      "required_pages_per_second_upper_estimate": 0.0017438036265595596
     },
     {
       "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": 29065.032828741965,
-      "required_pages_per_second_upper_estimate": 0.002259678458866199
+      "required_token_positions_per_second": 29088.374344211003,
+      "required_pages_per_second_upper_estimate": 0.0022614931590254402
     },
     {
       "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": 29065.032828741965,
-      "required_pages_per_second_upper_estimate": 0.3728150335262755
+      "required_token_positions_per_second": 29088.374344211003,
+      "required_pages_per_second_upper_estimate": 0.37311443342454564
     }
   ],
   "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": 0.05514428775782321,
-    "producer_p95_seconds": 0.15730371500012552,
-    "gpu_update_seconds": 0.9512461301147858,
-    "producer_batches_per_second": 18.13424455478858,
-    "gpu_batches_per_second": 1.0512526341414195,
-    "producer_headroom": 17.250129955297762,
+    "producer_batch_seconds": 0.049920898749999054,
+    "producer_p95_seconds": 0.15426751800032434,
+    "gpu_update_seconds": 0.9504828173906646,
+    "producer_batches_per_second": 20.031690635377814,
+    "gpu_batches_per_second": 1.0520968729821687,
+    "producer_headroom": 19.039777752212096,
     "sustainable": true,
-    "buffer_seconds": 60.87975232734629,
+    "buffer_seconds": 60.83090031300254,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -33,3 +33,4 @@
 {"step": 688300, "loss": 0.5151051878929138, "lr": 8e-05, "grad_norm": 0.18574555369965068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100748617566, "tokens_per_second": 154502.830518826, "data_wait_s": 1.603400005478761e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688400, "loss": 0.5832341909408569, "lr": 8e-05, "grad_norm": 0.21190185625883837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100763397324, "tokens_per_second": 155012.97129702306, "data_wait_s": 1.5720999726909213e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688500, "loss": 0.5517596006393433, "lr": 8e-05, "grad_norm": 0.18455588967613337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100778019023, "tokens_per_second": 153112.0184768985, "data_wait_s": 1.4700000065204222e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688600, "loss": 0.5771710872650146, "lr": 8e-05, "grad_norm": 0.2004392058658827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100792624872, "tokens_per_second": 153019.46159830692, "data_wait_s": 1.3870000202587107e-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 sha256:69edff6033d1193a260b0586c20c52e7dcbe6297b31a8e53240ae1effb9a83dd
-size 74295
+oid sha256:1a06027b2ab310d2566ec960de11aba0401c72259bd412ad6b1dc21d33b8b557
+size 76413