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": 76162.94955690732,
-      "required_pages_per_second_upper_estimate": 0.005269129572737414
+      "required_token_positions_per_second": 76719.48484152499,
+      "required_pages_per_second_upper_estimate": 0.005307631975067037
     },
     {
       "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": 9520.368694613415,
-      "required_pages_per_second_upper_estimate": 0.0039859795645290925
+      "required_token_positions_per_second": 9589.935605190623,
+      "required_pages_per_second_upper_estimate": 0.004015105777265483
     },
     {
       "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": 19040.73738922683,
-      "required_pages_per_second_upper_estimate": 0.003965683460163676
+      "required_token_positions_per_second": 19179.871210381247,
+      "required_pages_per_second_upper_estimate": 0.003994661365904534
     },
     {
       "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": 28561.106083840245,
-      "required_pages_per_second_upper_estimate": 0.00171219469944235
+      "required_token_positions_per_second": 28769.80681557187,
+      "required_pages_per_second_upper_estimate": 0.0017247059896420948
     },
     {
       "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": 28561.106083840245,
-      "required_pages_per_second_upper_estimate": 0.00222050037098959
+      "required_token_positions_per_second": 28769.80681557187,
+      "required_pages_per_second_upper_estimate": 0.0022367259349042198
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28561.106083840245,
-      "required_pages_per_second_upper_estimate": 0.3663512023170591
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28769.80681557187,
+      "required_pages_per_second_upper_estimate": 0.3690281912183254
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08082158564843667,
-    "producer_p95_seconds": 0.1727614250003171,
-    "gpu_update_seconds": 0.9680297366229498,
-    "producer_batches_per_second": 12.372932205882092,
-    "gpu_batches_per_second": 1.0330261170370458,
-    "producer_headroom": 11.977366304513655,
+    "producer_batch_seconds": 0.1464457547812188,
+    "producer_p95_seconds": 0.2032870080001885,
+    "gpu_update_seconds": 0.9610074957137118,
+    "producer_batches_per_second": 6.828466973958653,
+    "gpu_batches_per_second": 1.0405746099382187,
+    "producer_headroom": 6.562207946207793,
     "sustainable": true,
-    "buffer_seconds": 60.01784367062289,
+    "buffer_seconds": 60.54347222996385,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302429.45187419973,
-    "upload_bytes_per_second": 4764561.588323765,
+    "generation_bytes_per_second": 309349.9087667507,
+    "upload_bytes_per_second": 4781195.518718916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.89432017487997,
+    "estimated_full_shards_per_hour": 66.37928912403002,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 6180989,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3201266620008028,
+      "bytes": 6761235,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -64,3 +64,4 @@
 {"step": 726400, "loss": 0.5483208298683167, "lr": 5e-05, "grad_norm": 0.2387842533931959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106324813331, "tokens_per_second": 151435.70127050293, "data_wait_s": 1.1844000255223364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 726500, "loss": 0.5062057971954346, "lr": 5e-05, "grad_norm": 0.22825353047214517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106339248228, "tokens_per_second": 149008.91150842526, "data_wait_s": 9.424999916518573e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 726600, "loss": 0.5636150240898132, "lr": 5e-05, "grad_norm": 0.2252468873031572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106353808082, "tokens_per_second": 150484.9628767221, "data_wait_s": 1.3204000424593687e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 726700, "loss": 0.5367215275764465, "lr": 5e-05, "grad_norm": 0.2096011994271188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106368390154, "tokens_per_second": 150624.68540030325, "data_wait_s": 1.270099983230466e-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 sha256:745642a076088d7051fdf0a86d7be0e9bf4a8fa02dfef30a2ca5df368c17b187
-size 140736
+oid sha256:e45fab2701d29b10e24ecd5125f994439ced7c4d75223f310808e9351bd9c3ef
+size 142854