Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76107.51884380393,
-      "required_pages_per_second_upper_estimate": 0.006596806840637927
+      "required_token_positions_per_second": 75215.34044391656,
+      "required_pages_per_second_upper_estimate": 0.006519475078141171
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9513.439855475492,
-      "required_pages_per_second_upper_estimate": 0.003795053396950491
+      "required_token_positions_per_second": 9401.91755548957,
+      "required_pages_per_second_upper_estimate": 0.0037505654840791327
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19026.879710950983,
-      "required_pages_per_second_upper_estimate": 0.004401533759482543
+      "required_token_positions_per_second": 18803.83511097914,
+      "required_pages_per_second_upper_estimate": 0.004349936316729946
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28540.319566426475,
-      "required_pages_per_second_upper_estimate": 0.0017379557221110845
+      "required_token_positions_per_second": 28205.75266646871,
+      "required_pages_per_second_upper_estimate": 0.001717582353240521
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28540.319566426475,
-      "required_pages_per_second_upper_estimate": 0.002197112804884593
+      "required_token_positions_per_second": 28205.75266646871,
+      "required_pages_per_second_upper_estimate": 0.00217135691878539
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28540.319566426475,
-      "required_pages_per_second_upper_estimate": 0.35437525070994047
+      "required_token_positions_per_second": 28205.75266646871,
+      "required_pages_per_second_upper_estimate": 0.3502210495358495
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07777829463285002,
-    "producer_p95_seconds": 0.19753603199933423,
-    "gpu_update_seconds": 0.9687347731215961,
-    "producer_batches_per_second": 12.85705741840276,
-    "gpu_batches_per_second": 1.0322742898736428,
-    "producer_headroom": 12.455078601227733,
+    "producer_batch_seconds": 0.4093146664922216,
+    "producer_p95_seconds": 0.24857046699980856,
+    "gpu_update_seconds": 0.9802255705400207,
+    "producer_batches_per_second": 2.4431081558104966,
+    "gpu_batches_per_second": 1.0201733458647537,
+    "producer_headroom": 2.394797085920322,
     "sustainable": true,
-    "buffer_seconds": 60.061555933538955,
+    "buffer_seconds": 60.77398537348128,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326101.6330464016,
-    "upload_bytes_per_second": 5829918.304104281,
+    "generation_bytes_per_second": 276599.4469402284,
+    "upload_bytes_per_second": 5867042.411387903,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.97381919425999,
+    "estimated_full_shards_per_hour": 59.35180240779056,
     "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": 1.2238314549999814,
-      "bytes": 9984108,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.8533821090004494,
       "bytes": 2621863,
@@ -562,6 +556,12 @@
       "bytes": 6283978,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2377747039990936,
+      "bytes": 12765065,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -112,3 +112,4 @@
 {"step": 366200, "loss": 0.5273890234529972, "lr": 0.0003, "grad_norm": 0.13614282773534725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53601520104, "tokens_per_second": 148820.06353189057, "data_wait_s": 1.6324000171152875e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366300, "loss": 0.5714674312621355, "lr": 0.0003, "grad_norm": 0.1431296246208897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53616098050, "tokens_per_second": 149634.36440129453, "data_wait_s": 9.965999197447672e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 366400, "loss": 0.6159377992153168, "lr": 0.0003, "grad_norm": 0.31378930658636206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53630710851, "tokens_per_second": 146447.49119320296, "data_wait_s": 1.5109999367268756e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 366500, "loss": 0.6233857721090317, "lr": 0.0003, "grad_norm": 0.1690511630703369, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53645461592, "tokens_per_second": 150633.3828324391, "data_wait_s": 8.834998880047351e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f959da4a5eae7ea4f19dfc39fece2500517435b829d64f3fd670be3ae396c368
-size 243183
+oid sha256:cdb91148fa460ad2d15f8977f045093ed5909d51cde3c486319efb7382c35473
+size 245301