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": 74467.55649924783,
-      "required_pages_per_second_upper_estimate": 0.006454659061058366
+      "required_token_positions_per_second": 74798.28985180071,
+      "required_pages_per_second_upper_estimate": 0.006483326189821631
     },
     {
       "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": 9308.444562405979,
-      "required_pages_per_second_upper_estimate": 0.0037132777095923004
+      "required_token_positions_per_second": 9349.78623147509,
+      "required_pages_per_second_upper_estimate": 0.003729769519497004
     },
     {
       "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": 18616.889124811958,
-      "required_pages_per_second_upper_estimate": 0.004306689652967152
+      "required_token_positions_per_second": 18699.57246295018,
+      "required_pages_per_second_upper_estimate": 0.0043258169880683785
     },
     {
       "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": 27925.333687217935,
-      "required_pages_per_second_upper_estimate": 0.0017005063086488292
+      "required_token_positions_per_second": 28049.358694425267,
+      "required_pages_per_second_upper_estimate": 0.0017080587808788322
     },
     {
       "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": 27925.333687217935,
-      "required_pages_per_second_upper_estimate": 0.0021497694895132482
+      "required_token_positions_per_second": 28049.358694425267,
+      "required_pages_per_second_upper_estimate": 0.0021593172778912625
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27925.333687217935,
-      "required_pages_per_second_upper_estimate": 0.3467391843155064
+      "required_token_positions_per_second": 28049.358694425267,
+      "required_pages_per_second_upper_estimate": 0.34827915982002394
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13499817716416374,
-    "producer_p95_seconds": 0.44701500100200064,
-    "gpu_update_seconds": 0.990068742227962,
-    "producer_batches_per_second": 7.407507427185153,
-    "gpu_batches_per_second": 1.0100308769971764,
-    "producer_headroom": 7.333941561477491,
+    "producer_batch_seconds": 0.12629404483584494,
+    "producer_p95_seconds": 0.44597056900238385,
+    "gpu_update_seconds": 0.9856909849954952,
+    "producer_batches_per_second": 7.918029716284601,
+    "gpu_batches_per_second": 1.014516735186099,
+    "producer_headroom": 7.80473051026817,
     "sustainable": true,
-    "buffer_seconds": 60.394193275905685,
+    "buffer_seconds": 60.12715008472521,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478133.5785813136,
-    "upload_bytes_per_second": 4628037.193657025,
+    "generation_bytes_per_second": 494413.5793578813,
+    "upload_bytes_per_second": 4434315.267420349,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.59633558349186,
+    "estimated_full_shards_per_hour": 106.0896447711213,
     "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": 0.8202600639997399,
-      "bytes": 1256174,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.0791088629994192,
       "bytes": 10296860,
@@ -562,6 +556,12 @@
       "bytes": 7913091,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 4.708043279999401,
+      "bytes": 995949,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -412,3 +412,4 @@
 {"step": 196200, "loss": 0.5501316413283348, "lr": 0.0003, "grad_norm": 0.12685191393783454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28715614316, "tokens_per_second": 147621.3208553125, "data_wait_s": 1.270500069949776e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196300, "loss": 0.6403248496353626, "lr": 0.0003, "grad_norm": 0.17848715324458939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28730226039, "tokens_per_second": 148628.79923139638, "data_wait_s": 1.3712997315451503e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196400, "loss": 0.6060383096337318, "lr": 0.0003, "grad_norm": 0.15123548976325515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28744944508, "tokens_per_second": 150985.67238656757, "data_wait_s": 1.1547999747563154e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196500, "loss": 0.5728410333395004, "lr": 0.0003, "grad_norm": 0.13311416429812908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28759668846, "tokens_per_second": 150816.59064497318, "data_wait_s": 8.973998774308711e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6017549657a8476e503cd4550543ec6b00ed1f47eb0879a6e961273981f54445
-size 883281
+oid sha256:7e526f22a6d41974194dfdb4f635a3e912761545f63639cd0a7a4f83ea642c88
+size 885399