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": 76782.88124573114,
-      "required_pages_per_second_upper_estimate": 0.005312017885409943
+      "required_token_positions_per_second": 77302.7046460799,
+      "required_pages_per_second_upper_estimate": 0.005347980474402535
     },
     {
       "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": 9597.860155716393,
-      "required_pages_per_second_upper_estimate": 0.00401842362108719
+      "required_token_positions_per_second": 9662.838080759988,
+      "required_pages_per_second_upper_estimate": 0.00404562852140957
     },
     {
       "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": 19195.720311432786,
-      "required_pages_per_second_upper_estimate": 0.0039979623156846675
+      "required_token_positions_per_second": 19325.676161519976,
+      "required_pages_per_second_upper_estimate": 0.004025028692091596
     },
     {
       "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": 28793.58046714918,
-      "required_pages_per_second_upper_estimate": 0.0017261311837538907
+      "required_token_positions_per_second": 28988.51424227996,
+      "required_pages_per_second_upper_estimate": 0.0017378171659263406
     },
     {
       "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": 28793.58046714918,
-      "required_pages_per_second_upper_estimate": 0.0022385742317451132
+      "required_token_positions_per_second": 28988.51424227996,
+      "required_pages_per_second_upper_estimate": 0.0022537294753384703
     },
     {
       "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": 28793.58046714918,
-      "required_pages_per_second_upper_estimate": 0.3693331340946009
+      "required_token_positions_per_second": 28988.51424227996,
+      "required_pages_per_second_upper_estimate": 0.37183353525839796
     }
   ],
   "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": 0.06949543899995092,
+    "producer_batch_seconds": 0.08249604951564038,
     "producer_p95_seconds": 0.23346730900084367,
-    "gpu_update_seconds": 0.9602140321362194,
-    "producer_batches_per_second": 14.389433528158678,
-    "gpu_batches_per_second": 1.0414344787018655,
-    "producer_headroom": 13.81693598822935,
+    "gpu_update_seconds": 0.9537570559471857,
+    "producer_batches_per_second": 12.121792569114604,
+    "gpu_batches_per_second": 1.0484850348046861,
+    "producer_headroom": 11.561245193521218,
     "sustainable": true,
-    "buffer_seconds": 60.49348402458182,
+    "buffer_seconds": 60.0866945246727,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460172.93873254355,
+    "generation_bytes_per_second": 465986.23330700165,
     "upload_bytes_per_second": 5347256.208164543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.74240037424306,
+    "estimated_full_shards_per_hour": 99.98979806335008,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 579000, "loss": 0.5822747945785522, "lr": 0.0003, "grad_norm": 0.16719512047122234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84748502214, "tokens_per_second": 153617.60331346476, "data_wait_s": 1.343400072073564e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579100, "loss": 0.5804442167282104, "lr": 0.0003, "grad_norm": 0.16040496497450182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84763264848, "tokens_per_second": 153304.47354579673, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579200, "loss": 0.5281848311424255, "lr": 0.0003, "grad_norm": 0.16683918222658417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84777790977, "tokens_per_second": 150649.3625108852, "data_wait_s": 1.4092001947574317e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 579300, "loss": 0.5518209934234619, "lr": 0.0003, "grad_norm": 0.2045410055942124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84792385921, "tokens_per_second": 152927.73143455957, "data_wait_s": 1.0921001376118511e-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 sha256:438a610dcd564bc3bea53201bab5b5ec277f89598c4e4412ea4702a292b0a6ec
-size 471375
+oid sha256:47833c1f0439e893efac0eadca5af41ca5d003ff7d3776fe84ff87119c48b9b8
+size 473493