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": 76408.55179158223,
-      "required_pages_per_second_upper_estimate": 0.0052861209052079226
+      "required_token_positions_per_second": 77133.66646622385,
+      "required_pages_per_second_upper_estimate": 0.005336286020897478
     },
     {
       "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": 9551.068973947778,
-      "required_pages_per_second_upper_estimate": 0.003998833130391657
+      "required_token_positions_per_second": 9641.70830827798,
+      "required_pages_per_second_upper_estimate": 0.004036781926911179
     },
     {
       "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": 19102.137947895557,
-      "required_pages_per_second_upper_estimate": 0.003978471577292751
+      "required_token_positions_per_second": 19283.41661655596,
+      "required_pages_per_second_upper_estimate": 0.004016227143334736
     },
     {
       "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": 28653.206921843335,
-      "required_pages_per_second_upper_estimate": 0.0017177160040507983
+      "required_token_positions_per_second": 28925.12492483394,
+      "required_pages_per_second_upper_estimate": 0.0017340170731353353
     },
     {
       "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": 28653.206921843335,
-      "required_pages_per_second_upper_estimate": 0.002227660806035557
+      "required_token_positions_per_second": 28925.12492483394,
+      "required_pages_per_second_upper_estimate": 0.002248801234727183
     },
     {
       "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": 28653.206921843335,
-      "required_pages_per_second_upper_estimate": 0.36753257297678754
+      "required_token_positions_per_second": 28925.12492483394,
+      "required_pages_per_second_upper_estimate": 0.37102044515634663
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.7323661132734571,
-    "producer_p95_seconds": 0.2620542690019647,
-    "gpu_update_seconds": 0.9649181704307929,
-    "producer_batches_per_second": 1.3654372886401032,
-    "gpu_batches_per_second": 1.0363573105412085,
-    "producer_headroom": 1.3175352503925908,
+    "producer_batch_seconds": 0.06794957001562807,
+    "producer_p95_seconds": 0.2471458900035941,
+    "gpu_update_seconds": 0.9558472114415155,
+    "producer_batches_per_second": 14.716796585614963,
+    "gpu_batches_per_second": 1.0461923077558573,
+    "producer_headroom": 14.067008977712078,
     "sustainable": true,
-    "buffer_seconds": 60.78984473713995,
+    "buffer_seconds": 60.21837432081548,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282586.811917692,
-    "upload_bytes_per_second": 5419462.490743002,
+    "generation_bytes_per_second": 281509.64892443246,
+    "upload_bytes_per_second": 5568074.546776141,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.63655155323096,
+    "estimated_full_shards_per_hour": 60.4054174499486,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.0856502950009599,
-      "bytes": 2667766,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.1392225029994734,
       "bytes": 10365382,
@@ -562,6 +556,12 @@
       "bytes": 2461242,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.3424353250011336,
+      "bytes": 15776988,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -235,3 +235,4 @@
 {"step": 708500, "loss": 0.5990826487541199, "lr": 8e-05, "grad_norm": 0.21880586457356885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103704404070, "tokens_per_second": 153765.63286826626, "data_wait_s": 1.557999712531455e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708600, "loss": 0.5554506182670593, "lr": 8e-05, "grad_norm": 0.22612946778818444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103719089001, "tokens_per_second": 137928.13761065842, "data_wait_s": 1.6326001059496775e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708700, "loss": 0.48246949911117554, "lr": 8e-05, "grad_norm": 0.21902118633597523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103733824275, "tokens_per_second": 154016.21070350928, "data_wait_s": 1.7959999240702018e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 708800, "loss": 0.5057194232940674, "lr": 8e-05, "grad_norm": 0.20662790801542463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103748414967, "tokens_per_second": 151814.69154172594, "data_wait_s": 1.5845002053538337e-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:fa712d3a77d2ff2a6e3241116907f4dd63d11330b48277d3e04df2eebcd06b82
-size 505263
+oid sha256:ad91f312b1267b2d34e20e50c19abd53248298ccaefe5ad00905e2abeab36703
+size 507381