Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75512.56195927144,
-      "required_pages_per_second_upper_estimate": 0.006545237485791088
+      "required_token_positions_per_second": 74389.06575289226,
+      "required_pages_per_second_upper_estimate": 0.006447855682097243
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9439.07024490893,
-      "required_pages_per_second_upper_estimate": 0.003765386247370723
+      "required_token_positions_per_second": 9298.633219111533,
+      "required_pages_per_second_upper_estimate": 0.003709363818059491
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18878.14048981786,
-      "required_pages_per_second_upper_estimate": 0.00436712555839428
+      "required_token_positions_per_second": 18597.266438223065,
+      "required_pages_per_second_upper_estimate": 0.004302150289772301
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28317.21073472679,
-      "required_pages_per_second_upper_estimate": 0.0017243695648221477
+      "required_token_positions_per_second": 27895.8996573346,
+      "required_pages_per_second_upper_estimate": 0.001698713930657354
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28317.21073472679,
-      "required_pages_per_second_upper_estimate": 0.002179937269415577
+      "required_token_positions_per_second": 27895.8996573346,
+      "required_pages_per_second_upper_estimate": 0.0021475035764142196
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28317.21073472679,
-      "required_pages_per_second_upper_estimate": 0.3516049857174564
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27895.8996573346,
+      "required_pages_per_second_upper_estimate": 0.34637371217371643
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07966900380480979,
-    "producer_p95_seconds": 0.22503182099899277,
-    "gpu_update_seconds": 0.9763673498426134,
-    "producer_batches_per_second": 12.551933025923287,
-    "gpu_batches_per_second": 1.024204670671542,
-    "producer_headroom": 12.255297583922696,
+    "producer_batch_seconds": 0.22863940589857634,
+    "producer_p95_seconds": 0.2969011440000031,
+    "gpu_update_seconds": 0.991113401597377,
+    "producer_batches_per_second": 4.373699258314188,
+    "gpu_batches_per_second": 1.0089662781154007,
+    "producer_headroom": 4.3348319494717,
     "sustainable": true,
-    "buffer_seconds": 60.53477569024203,
+    "buffer_seconds": 60.45791749744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 634204.3550508835,
+    "generation_bytes_per_second": 574167.803898092,
     "upload_bytes_per_second": 5722526.054790809,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.04274553606452,
+    "estimated_full_shards_per_hour": 61.60152238706145,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -170,3 +170,4 @@
 {"step": 372000, "loss": 0.5794756803661585, "lr": 0.0003, "grad_norm": 0.1618646699010155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54449914420, "tokens_per_second": 150231.908433979, "data_wait_s": 8.095001248875633e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 372100, "loss": 0.6335708033293486, "lr": 0.0003, "grad_norm": 0.18131454470469277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54464514533, "tokens_per_second": 148717.74400200957, "data_wait_s": 9.993000276153907e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 372200, "loss": 0.5570940934121609, "lr": 0.0003, "grad_norm": 0.14605080317111024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54479184826, "tokens_per_second": 148065.656737445, "data_wait_s": 7.60500188334845e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 372300, "loss": 0.5817138999700546, "lr": 0.0003, "grad_norm": 0.17083567174265302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54493951271, "tokens_per_second": 150660.22467692784, "data_wait_s": 1.0944000678136945e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80a64ac2def6958f79ae98d1905736ca4c0e9194ab6711dc44a3c10eb917f4a3
-size 366810
+oid sha256:aecabd9d1f5b2e13d010c66de37968be0a124bd532adee21a76f9264f9f3dd90
+size 368928