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": 77387.57019974591,
-      "required_pages_per_second_upper_estimate": 0.005353851670320363
+      "required_token_positions_per_second": 76395.03562198069,
+      "required_pages_per_second_upper_estimate": 0.005285185825233053
     },
     {
       "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": 9673.446274968239,
-      "required_pages_per_second_upper_estimate": 0.0040500699508002795
+      "required_token_positions_per_second": 9549.379452747586,
+      "required_pages_per_second_upper_estimate": 0.003998125763146351
     },
     {
       "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": 19346.892549936478,
-      "required_pages_per_second_upper_estimate": 0.004029447506284965
+      "required_token_positions_per_second": 19098.75890549517,
+      "required_pages_per_second_upper_estimate": 0.0039777678118720905
     },
     {
       "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": 29020.338824904717,
-      "required_pages_per_second_upper_estimate": 0.0017397250010613707
+      "required_token_positions_per_second": 28648.138358242755,
+      "required_pages_per_second_upper_estimate": 0.0017174121514021922
     },
     {
       "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": 29020.338824904717,
-      "required_pages_per_second_upper_estimate": 0.002256203696656
+      "required_token_positions_per_second": 28648.138358242755,
+      "required_pages_per_second_upper_estimate": 0.0022272667475098673
     },
     {
       "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": 29020.338824904717,
-      "required_pages_per_second_upper_estimate": 0.37224174683373373
+      "required_token_positions_per_second": 28648.138358242755,
+      "required_pages_per_second_upper_estimate": 0.3674675588851189
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.15990689707047068,
-    "producer_p95_seconds": 0.370896681000886,
-    "gpu_update_seconds": 0.9527111370689098,
-    "producer_batches_per_second": 6.253638950665786,
-    "gpu_batches_per_second": 1.0496360975442967,
-    "producer_headroom": 5.957911475507224,
+    "producer_batch_seconds": 0.13329735316432334,
+    "producer_p95_seconds": 0.36278332399524515,
+    "gpu_update_seconds": 0.9650888882992639,
+    "producer_batches_per_second": 7.50202443080203,
+    "gpu_batches_per_second": 1.0361739857582015,
+    "producer_headroom": 7.240120417916649,
     "sustainable": true,
-    "buffer_seconds": 60.020801635341314,
+    "buffer_seconds": 60.800599962853624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298837.00476091955,
-    "upload_bytes_per_second": 5392553.106141569,
+    "generation_bytes_per_second": 298978.01718284166,
+    "upload_bytes_per_second": 5442207.661463906,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.12346465225878,
+    "estimated_full_shards_per_hour": 64.15372263540208,
     "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": [
-    {
-      "seconds": 1.743612624995876,
-      "bytes": 7907888,
-      "pages": 23,
-      "shards": 3
-    },
     {
       "seconds": 2.2645728169954964,
       "bytes": 6235357,
@@ -562,6 +556,12 @@
       "bytes": 2339484,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.3525367339971126,
+      "bytes": 9731842,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -427,3 +427,4 @@
 {"step": 482900, "loss": 0.5762990117073059, "lr": 0.0003, "grad_norm": 0.1542558287123174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70680327522, "tokens_per_second": 152236.4964049248, "data_wait_s": 1.4051998732611537e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 483000, "loss": 0.5855228304862976, "lr": 0.0003, "grad_norm": 0.14452549252851202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70695102401, "tokens_per_second": 153096.64240842784, "data_wait_s": 1.3914002920500934e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 483100, "loss": 0.5663217306137085, "lr": 0.0003, "grad_norm": 0.17280523039607762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70709519840, "tokens_per_second": 149986.1279342234, "data_wait_s": 1.4061006368137896e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 483200, "loss": 0.5441246032714844, "lr": 0.0003, "grad_norm": 0.163929241214397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70724007623, "tokens_per_second": 149481.46959061088, "data_wait_s": 9.726994903758168e-06, "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 sha256:1446b0b8136c27787c9cbd9b97b56372bf85ed35f096f70f5cdac4a79375090c
-size 915051
+oid sha256:e87798e28fc1e3e572f5696200ff56ddf94c15bf75625a121e994005ac73501c
+size 917169