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": 77334.85143249403,
-      "required_pages_per_second_upper_estimate": 0.005350204463677485
+      "required_token_positions_per_second": 77313.21059275793,
+      "required_pages_per_second_upper_estimate": 0.005348707300170875
     },
     {
       "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": 9666.856429061754,
-      "required_pages_per_second_upper_estimate": 0.004047310919930865
+      "required_token_positions_per_second": 9664.15132409474,
+      "required_pages_per_second_upper_estimate": 0.004046178348970192
     },
     {
       "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": 19333.71285812351,
-      "required_pages_per_second_upper_estimate": 0.004026702524052169
+      "required_token_positions_per_second": 19328.30264818948,
+      "required_pages_per_second_upper_estimate": 0.004025575719999742
     },
     {
       "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": 29000.56928718526,
-      "required_pages_per_second_upper_estimate": 0.0017385398474614256
+      "required_token_positions_per_second": 28992.453972284224,
+      "required_pages_per_second_upper_estimate": 0.0017380533467244769
     },
     {
       "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": 29000.56928718526,
-      "required_pages_per_second_upper_estimate": 0.002254666702055318
+      "required_token_positions_per_second": 28992.453972284224,
+      "required_pages_per_second_upper_estimate": 0.0022540357720172606
     },
     {
       "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": 29000.56928718526,
-      "required_pages_per_second_upper_estimate": 0.37198816443074434
+      "required_token_positions_per_second": 28992.453972284224,
+      "required_pages_per_second_upper_estimate": 0.3718840698847401
     }
   ],
   "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": 0.10632708937521329,
-    "producer_p95_seconds": 0.32807919399783714,
-    "gpu_update_seconds": 0.9533605953113847,
-    "producer_batches_per_second": 9.404940978598042,
-    "gpu_batches_per_second": 1.0489210535006244,
-    "producer_headroom": 8.966300130224667,
+    "producer_batch_seconds": 0.14006221950006648,
+    "producer_p95_seconds": 0.3495665270020254,
+    "gpu_update_seconds": 0.9536274516958974,
+    "producer_batches_per_second": 7.139684088752609,
+    "gpu_batches_per_second": 1.0486275308262523,
+    "producer_headroom": 6.808598743470895,
     "sustainable": true,
-    "buffer_seconds": 60.061717504617235,
+    "buffer_seconds": 60.07852945684154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329689.041779404,
+    "generation_bytes_per_second": 349333.6393321442,
     "upload_bytes_per_second": 5121752.382386376,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.74359359776105,
+    "estimated_full_shards_per_hour": 74.95886692975277,
     "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": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -383,3 +383,4 @@
 {"step": 963300, "loss": 0.4876033067703247, "lr": 3e-05, "grad_norm": 0.3265877991015855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141000884630, "tokens_per_second": 152203.69996306024, "data_wait_s": 8.148002962116152e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 963400, "loss": 0.5462270975112915, "lr": 3e-05, "grad_norm": 0.34679611102436675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141015623594, "tokens_per_second": 153549.73405511674, "data_wait_s": 1.4596997061744332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 963500, "loss": 0.5195978283882141, "lr": 3e-05, "grad_norm": 0.34887611637436744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141030197919, "tokens_per_second": 151459.24333926337, "data_wait_s": 1.9879000319633633e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 963600, "loss": 0.5554284453392029, "lr": 3e-05, "grad_norm": 0.33351172778197596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141044863094, "tokens_per_second": 152538.94204790838, "data_wait_s": 1.1431999155320227e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7751baa7345871890b9a1de1f83795275de8fd65151b973234475dff6cf7950e
-size 821076
+oid sha256:a278e4b0975eb029f00d7303e21968c7165b7846ab28014c8c91aaca9c25fb5b
+size 823194