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": 76484.75841878729,
-      "required_pages_per_second_upper_estimate": 0.005291393056501705
+      "required_token_positions_per_second": 77009.97995738935,
+      "required_pages_per_second_upper_estimate": 0.005327729101224067
     },
     {
       "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": 9560.59480234841,
-      "required_pages_per_second_upper_estimate": 0.004002821395821084
+      "required_token_positions_per_second": 9626.24749467367,
+      "required_pages_per_second_upper_estimate": 0.004030308807113555
     },
     {
       "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": 19121.18960469682,
-      "required_pages_per_second_upper_estimate": 0.003982439534978477
+      "required_token_positions_per_second": 19252.49498934734,
+      "required_pages_per_second_upper_estimate": 0.004009786983845327
     },
     {
       "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": 28681.78440704523,
-      "required_pages_per_second_upper_estimate": 0.0017194291806533586
+      "required_token_positions_per_second": 28878.742484021004,
+      "required_pages_per_second_upper_estimate": 0.001731236516630486
     },
     {
       "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": 28681.78440704523,
-      "required_pages_per_second_upper_estimate": 0.0022298825798109313
+      "required_token_positions_per_second": 28878.742484021004,
+      "required_pages_per_second_upper_estimate": 0.0022451952039687735
     },
     {
       "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": 28681.78440704523,
-      "required_pages_per_second_upper_estimate": 0.3678991342728445
+      "required_token_positions_per_second": 28878.742484021004,
+      "required_pages_per_second_upper_estimate": 0.37042550100718313
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.047388032406203706,
-    "producer_p95_seconds": 0.1575497110006836,
-    "gpu_update_seconds": 0.9639567611145108,
-    "producer_batches_per_second": 21.102374359588037,
-    "gpu_batches_per_second": 1.0373909290742633,
-    "producer_headroom": 20.34177643949438,
+    "producer_batch_seconds": 0.07044797614843645,
+    "producer_p95_seconds": 0.15542690500024037,
+    "gpu_update_seconds": 0.9573824073294745,
+    "producer_batches_per_second": 14.194871941998214,
+    "gpu_batches_per_second": 1.0445147021130283,
+    "producer_headroom": 13.589920671563863,
     "sustainable": true,
-    "buffer_seconds": 60.72927595021418,
+    "buffer_seconds": 60.315091661756895,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 175,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357440.0075367105,
-    "upload_bytes_per_second": 323547.02401647536,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 792932.9292294211,
+    "upload_bytes_per_second": 1604185.3993816108,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.69830483985888,
+    "estimated_full_shards_per_hour": 85.07247403937328,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 1003408,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.010816134000379,
+      "bytes": 5780671,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -41,3 +41,4 @@
 {"step": 724100, "loss": 0.5465595126152039, "lr": 5e-05, "grad_norm": 0.2197309375628237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105988236583, "tokens_per_second": 151372.11473878397, "data_wait_s": 2.0450999727472663e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724200, "loss": 0.5930379033088684, "lr": 5e-05, "grad_norm": 0.23605915453086354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106003060102, "tokens_per_second": 152961.70009792663, "data_wait_s": 1.2775999493896961e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724300, "loss": 0.4928969442844391, "lr": 5e-05, "grad_norm": 0.21268350994343688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106017674194, "tokens_per_second": 151715.12888789756, "data_wait_s": 1.1274999451416079e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 724400, "loss": 0.4595949649810791, "lr": 5e-05, "grad_norm": 0.21529503899523042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106032156398, "tokens_per_second": 115167.31695636589, "data_wait_s": 1.4557999747921713e-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 sha256:2084258eb9433368b9f9aeb52df483a0abe745c82d46fec73dc30c029cdd7aa5
-size 91239
+oid sha256:266b622d04a649cf89bbcf24c95912e3ee1c0b7b8dc667efaf902035ea1fd044
+size 93357