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": 76348.81847415636,
-      "required_pages_per_second_upper_estimate": 0.005281988415708009
+      "required_token_positions_per_second": 76173.31822422586,
+      "required_pages_per_second_upper_estimate": 0.005269846901201134
     },
     {
       "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": 9543.602309269545,
-      "required_pages_per_second_upper_estimate": 0.0039957069938125695
+      "required_token_positions_per_second": 9521.664778028233,
+      "required_pages_per_second_upper_estimate": 0.003986522207589578
     },
     {
       "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": 19087.20461853909,
-      "required_pages_per_second_upper_estimate": 0.003975361358606177
+      "required_token_positions_per_second": 19043.329556056466,
+      "required_pages_per_second_upper_estimate": 0.0039662233401542525
     },
     {
       "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": 28630.806927808637,
-      "required_pages_per_second_upper_estimate": 0.0017163731586112287
+      "required_token_positions_per_second": 28564.9943340847,
+      "required_pages_per_second_upper_estimate": 0.0017124277941074906
     },
     {
       "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": 28630.806927808637,
-      "required_pages_per_second_upper_estimate": 0.0022259193050265203
+      "required_token_positions_per_second": 28564.9943340847,
+      "required_pages_per_second_upper_estimate": 0.002220802665343491
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28630.806927808637,
-      "required_pages_per_second_upper_estimate": 0.3672452499045502
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28564.9943340847,
+      "required_pages_per_second_upper_estimate": 0.3664010766163171
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08223363003906314,
-    "producer_p95_seconds": 0.1606245080001827,
-    "gpu_update_seconds": 0.9656730971541688,
-    "producer_batches_per_second": 12.16047497264773,
-    "gpu_batches_per_second": 1.0355471255717823,
-    "producer_headroom": 11.74304352970249,
+    "producer_batch_seconds": 0.06641513317970293,
+    "producer_p95_seconds": 0.1675205020001158,
+    "gpu_update_seconds": 0.9678979689839984,
+    "producer_batches_per_second": 15.056809376475195,
+    "gpu_batches_per_second": 1.0331667510881328,
+    "producer_headroom": 14.573455214869565,
     "sustainable": true,
-    "buffer_seconds": 60.83740512071264,
+    "buffer_seconds": 60.0096740770079,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393120.72105182114,
-    "upload_bytes_per_second": 5685116.03939964,
+    "generation_bytes_per_second": 425041.0941170548,
+    "upload_bytes_per_second": 5624807.609069092,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.35455535570121,
+    "estimated_full_shards_per_hour": 91.20392434724553,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 6490657,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8932356570003321,
+      "bytes": 2377409,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -46,3 +46,4 @@
 {"step": 561700, "loss": 0.582870602607727, "lr": 0.0003, "grad_norm": 0.1587683086837042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82215218311, "tokens_per_second": 151894.4562989931, "data_wait_s": 1.654700008657528e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561800, "loss": 0.567470908164978, "lr": 0.0003, "grad_norm": 0.17873162162093043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82229882507, "tokens_per_second": 152559.57835867302, "data_wait_s": 1.4860999726806767e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561900, "loss": 0.5914993286132812, "lr": 0.0003, "grad_norm": 0.14266866861507754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82244640620, "tokens_per_second": 153630.32665940447, "data_wait_s": 1.4713000382471364e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 562000, "loss": 0.5662038326263428, "lr": 0.0003, "grad_norm": 0.1866444866498162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82259344566, "tokens_per_second": 153271.57740873244, "data_wait_s": 1.495099968451541e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0db134c71444dbe174e1d37dc1001213e249c1043900a72b3eab3c981de5e417
-size 102612
+oid sha256:9487ffa4dfe61079f56f0d014fdd645a91ece134a17e52acc1a587c702dbe24a
+size 104730