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": 76927.88539731651,
-      "required_pages_per_second_upper_estimate": 0.0053220496090726046
+      "required_token_positions_per_second": 77103.06102905466,
+      "required_pages_per_second_upper_estimate": 0.005334168665739715
     },
     {
       "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": 9615.985674664564,
-      "required_pages_per_second_upper_estimate": 0.004026012397366912
+      "required_token_positions_per_second": 9637.882628631833,
+      "required_pages_per_second_upper_estimate": 0.004035180194732611
     },
     {
       "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": 19231.971349329127,
-      "required_pages_per_second_upper_estimate": 0.004005512450874317
+      "required_token_positions_per_second": 19275.765257263665,
+      "required_pages_per_second_upper_estimate": 0.004014633566974064
     },
     {
       "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": 28847.95702399369,
-      "required_pages_per_second_upper_estimate": 0.0017293909753085242
+      "required_token_positions_per_second": 28913.647885895498,
+      "required_pages_per_second_upper_estimate": 0.0017333290421753993
     },
     {
       "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": 28847.95702399369,
-      "required_pages_per_second_upper_estimate": 0.0022428017698625778
+      "required_token_positions_per_second": 28913.647885895498,
+      "required_pages_per_second_upper_estimate": 0.002247908945431878
     },
     {
       "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": 28847.95702399369,
-      "required_pages_per_second_upper_estimate": 0.3700306181808044
+      "required_token_positions_per_second": 28913.647885895498,
+      "required_pages_per_second_upper_estimate": 0.3708732300239286
     }
   ],
   "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": 0.3350716874917907,
-    "producer_p95_seconds": 0.32063350600219565,
-    "gpu_update_seconds": 0.9584040900020875,
-    "producer_batches_per_second": 2.9844359799110154,
-    "gpu_batches_per_second": 1.0434012233794014,
-    "producer_headroom": 2.860295649496105,
+    "producer_batch_seconds": 0.11831319244492988,
+    "producer_p95_seconds": 0.3144909490001737,
+    "gpu_update_seconds": 0.9562266272699234,
+    "producer_batches_per_second": 8.452142819706776,
+    "gpu_batches_per_second": 1.0457771949470305,
+    "producer_headroom": 8.08216402169191,
     "sustainable": true,
-    "buffer_seconds": 60.37945767013151,
+    "buffer_seconds": 60.242277518005174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314241.86256631423,
+    "generation_bytes_per_second": 355518.7531721854,
     "upload_bytes_per_second": 5233381.266217245,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.4289885305602,
+    "estimated_full_shards_per_hour": 76.28604837774441,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -350,3 +350,4 @@
 {"step": 475200, "loss": 0.6070004105567932, "lr": 0.0003, "grad_norm": 0.15388277254052188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69554224058, "tokens_per_second": 134917.34085297797, "data_wait_s": 8.978000550996512e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 475300, "loss": 0.5609617233276367, "lr": 0.0003, "grad_norm": 0.17542387447329194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69568895568, "tokens_per_second": 153043.73075176837, "data_wait_s": 1.012799475574866e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475400, "loss": 0.5558681488037109, "lr": 0.0003, "grad_norm": 0.1455002630257173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69583454500, "tokens_per_second": 151211.05860750322, "data_wait_s": 1.4534998626913875e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 475500, "loss": 0.6086797714233398, "lr": 0.0003, "grad_norm": 0.16874298172772706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69597926333, "tokens_per_second": 150589.9090848339, "data_wait_s": 1.1281001206953079e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea6b2022efe6ff3529566f25f6beb48649c99fe898c05d8785ebc2adcc56b39f
-size 751182
+oid sha256:49453e7c25000d47e391fb8afe89467cd07681a630ec41e5418106ee31bd151b
+size 753300