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": 77187.88083248197,
-      "required_pages_per_second_upper_estimate": 0.005340036696549875
+      "required_token_positions_per_second": 77028.58290861282,
+      "required_pages_per_second_upper_estimate": 0.0053290160965545
     },
     {
       "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": 9648.485104060246,
-      "required_pages_per_second_upper_estimate": 0.004039619229789624
+      "required_token_positions_per_second": 9628.572863576603,
+      "required_pages_per_second_upper_estimate": 0.004031282390513988
     },
     {
       "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": 19296.970208120492,
-      "required_pages_per_second_upper_estimate": 0.004019049999025382
+      "required_token_positions_per_second": 19257.145727153205,
+      "required_pages_per_second_upper_estimate": 0.004010755609882085
     },
     {
       "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": 28945.455312180737,
-      "required_pages_per_second_upper_estimate": 0.0017352358488140715
+      "required_token_positions_per_second": 28885.718590729808,
+      "required_pages_per_second_upper_estimate": 0.0017316547235757807
     },
     {
       "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": 28945.455312180737,
-      "required_pages_per_second_upper_estimate": 0.002250381832919473
+      "required_token_positions_per_second": 28885.718590729808,
+      "required_pages_per_second_upper_estimate": 0.0022457375655807322
     },
     {
       "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": 28945.455312180737,
-      "required_pages_per_second_upper_estimate": 0.37128122153616216
+      "required_token_positions_per_second": 28885.718590729808,
+      "required_pages_per_second_upper_estimate": 0.37051498301368385
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1950242052733131,
-    "producer_p95_seconds": 0.23554617099944153,
-    "gpu_update_seconds": 0.9551758540956602,
-    "producer_batches_per_second": 5.127568645125708,
-    "gpu_batches_per_second": 1.0469276371593148,
-    "producer_headroom": 4.8977297600420755,
+    "producer_batch_seconds": 0.08128841430456646,
+    "producer_p95_seconds": 0.22643614199841977,
+    "gpu_update_seconds": 0.9571511926614481,
+    "producer_batches_per_second": 12.301876085972857,
+    "gpu_batches_per_second": 1.0447670207873918,
+    "producer_headroom": 11.774755367662268,
     "sustainable": true,
-    "buffer_seconds": 60.176078808026595,
+    "buffer_seconds": 60.30052513767123,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374615.9095459555,
+    "generation_bytes_per_second": 415932.69936200517,
     "upload_bytes_per_second": 5504521.933352885,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.38385357650755,
+    "estimated_full_shards_per_hour": 89.24947486538998,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -163,3 +163,4 @@
 {"step": 501300, "loss": 0.6153491139411926, "lr": 0.0003, "grad_norm": 0.15550178698734593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73373699778, "tokens_per_second": 153723.37926457796, "data_wait_s": 1.4983001165091991e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501400, "loss": 0.6103640794754028, "lr": 0.0003, "grad_norm": 0.15216060407855822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73388345881, "tokens_per_second": 152953.89000887508, "data_wait_s": 1.0103998647537082e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501500, "loss": 0.6360791325569153, "lr": 0.0003, "grad_norm": 0.1606787510080989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73403039239, "tokens_per_second": 152718.59530307463, "data_wait_s": 1.2568998499773443e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 501600, "loss": 0.5874972939491272, "lr": 0.0003, "grad_norm": 0.14712835035756988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73417566943, "tokens_per_second": 151537.9510395918, "data_wait_s": 1.1756997992051765e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ca71cc4ee8ed9abc6765eea21731064fcc77a7e0c032dcc34de3394fb6d567d
-size 351984
+oid sha256:eb653c2277fa9905c120ed01c8a94afd0650238bf376ba629516780f44db441d
+size 354102