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": 76092.44395755992,
-      "required_pages_per_second_upper_estimate": 0.006595500187705246
+      "required_token_positions_per_second": 76108.73656085691,
+      "required_pages_per_second_upper_estimate": 0.006596912389265803
     },
     {
       "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": 9511.55549469499,
-      "required_pages_per_second_upper_estimate": 0.003794301697261445
+      "required_token_positions_per_second": 9513.592070107114,
+      "required_pages_per_second_upper_estimate": 0.003795114117642857
     },
     {
       "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": 19023.11098938998,
-      "required_pages_per_second_upper_estimate": 0.004400661932076639
+      "required_token_positions_per_second": 19027.184140214227,
+      "required_pages_per_second_upper_estimate": 0.004401604183834822
     },
     {
       "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": 28534.666484084973,
-      "required_pages_per_second_upper_estimate": 0.0017376114790559178
+      "required_token_positions_per_second": 28540.776210321343,
+      "required_pages_per_second_upper_estimate": 0.0017379835293284565
     },
     {
       "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": 28534.666484084973,
-      "required_pages_per_second_upper_estimate": 0.0021966776149572108
+      "required_token_positions_per_second": 28540.776210321343,
+      "required_pages_per_second_upper_estimate": 0.0021971479585956915
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28534.666484084973,
-      "required_pages_per_second_upper_estimate": 0.35430505834690856
+      "required_token_positions_per_second": 28540.776210321343,
+      "required_pages_per_second_upper_estimate": 0.35438092069882593
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.25723577958595456,
-    "producer_p95_seconds": 0.31099102499865694,
-    "gpu_update_seconds": 0.9689266918686607,
-    "producer_batches_per_second": 3.8874840879818313,
-    "gpu_batches_per_second": 1.0320698236431196,
-    "producer_headroom": 3.7666870970602933,
+    "producer_batch_seconds": 0.09583441909370549,
+    "producer_p95_seconds": 0.3272251869966567,
+    "gpu_update_seconds": 0.9687192736545658,
+    "producer_batches_per_second": 10.434664387355598,
+    "gpu_batches_per_second": 1.03229080621822,
+    "producer_headroom": 10.108260506148278,
     "sustainable": true,
-    "buffer_seconds": 60.07345489585696,
+    "buffer_seconds": 60.06059496658308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483114.6669468739,
+    "generation_bytes_per_second": 502311.655093655,
     "upload_bytes_per_second": 4552632.234319538,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.66516119293844,
+    "estimated_full_shards_per_hour": 107.78438796622503,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -249,3 +249,4 @@
 {"step": 64900, "loss": 0.6011261399835348, "lr": 0.0003, "grad_norm": 0.13736492854896148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9498208077, "tokens_per_second": 148070.96434491608, "data_wait_s": 1.1184001778019592e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65000, "loss": 0.6491995714604855, "lr": 0.0003, "grad_norm": 0.12935625010639407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9512775355, "tokens_per_second": 148919.8180950165, "data_wait_s": 1.1274001735728234e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65100, "loss": 0.5611817017197609, "lr": 0.0003, "grad_norm": 0.12788663244318282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9527447309, "tokens_per_second": 133901.81024125547, "data_wait_s": 9.081999451154843e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 65200, "loss": 0.5868961215019226, "lr": 0.0003, "grad_norm": 0.13932515037847346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9542147796, "tokens_per_second": 149279.75819578778, "data_wait_s": 1.1594001989578828e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84c6680035e3f037694f16c8b1c51cd32e1bf5bc503af12b2002ab4009a77597
-size 535698
+oid sha256:549555e87fff747f472c6cdd231c3da095c2aa228d193438976ed727d087f4e1
+size 537816