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": 75599.07269866343,
-      "required_pages_per_second_upper_estimate": 0.006552736017422121
+      "required_token_positions_per_second": 75951.58535364216,
+      "required_pages_per_second_upper_estimate": 0.006583290920920522
     },
     {
       "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": 9449.884087332928,
-      "required_pages_per_second_upper_estimate": 0.003769700050795009
+      "required_token_positions_per_second": 9493.94816920527,
+      "required_pages_per_second_upper_estimate": 0.00378727787187062
     },
     {
       "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": 18899.768174665856,
-      "required_pages_per_second_upper_estimate": 0.004372128742649611
+      "required_token_positions_per_second": 18987.89633841054,
+      "required_pages_per_second_upper_estimate": 0.004392515642329235
     },
     {
       "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": 28349.652261998788,
-      "required_pages_per_second_upper_estimate": 0.001726345083625473
+      "required_token_positions_per_second": 28481.844507615813,
+      "required_pages_per_second_upper_estimate": 0.0017343948978244383
     },
     {
       "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": 28349.652261998788,
-      "required_pages_per_second_upper_estimate": 0.0021824347079888702
+      "required_token_positions_per_second": 28481.844507615813,
+      "required_pages_per_second_upper_estimate": 0.0021926112329880257
     },
     {
       "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": 28349.652261998788,
-      "required_pages_per_second_upper_estimate": 0.3520078009113673
+      "required_token_positions_per_second": 28481.844507615813,
+      "required_pages_per_second_upper_estimate": 0.35364918618294466
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09934278121853879,
-    "producer_p95_seconds": 0.2410791540005448,
-    "gpu_update_seconds": 0.9752500575487018,
-    "producer_batches_per_second": 10.066156672221148,
-    "gpu_batches_per_second": 1.0253780476706738,
-    "producer_headroom": 9.817019873877923,
+    "producer_batch_seconds": 0.08484108418741698,
+    "producer_p95_seconds": 0.23071563499979675,
+    "gpu_update_seconds": 0.9707236479227022,
+    "producer_batches_per_second": 11.786742349861587,
+    "gpu_batches_per_second": 1.0301593065543913,
+    "producer_headroom": 11.441669530982642,
     "sustainable": true,
-    "buffer_seconds": 60.46550356801951,
+    "buffer_seconds": 60.18486617120753,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339368.83086022496,
-    "upload_bytes_per_second": 4037676.412633108,
+    "generation_bytes_per_second": 338929.477969675,
+    "upload_bytes_per_second": 4029110.3229128295,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.820651000548,
+    "estimated_full_shards_per_hour": 72.72637609784782,
     "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": 0.8108393100001194,
-      "bytes": 1203088,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1118799440000657,
       "bytes": 10497270,
@@ -562,6 +556,12 @@
       "bytes": 12103363,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.0281697059999715,
+      "bytes": 1191797,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -111,3 +111,4 @@
 {"step": 326100, "loss": 0.5768279675394297, "lr": 0.0003, "grad_norm": 0.15366335743806186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47732037859, "tokens_per_second": 134682.01316421214, "data_wait_s": 1.824300125008449e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 326200, "loss": 0.5857859924435616, "lr": 0.0003, "grad_norm": 0.14807730818720793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47746643582, "tokens_per_second": 148480.21576413972, "data_wait_s": 1.8270999134983867e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 326300, "loss": 0.5941286776214838, "lr": 0.0003, "grad_norm": 0.1455359417592417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47761346946, "tokens_per_second": 149949.39670711057, "data_wait_s": 1.2856999092036858e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 326400, "loss": 0.6128942668437958, "lr": 0.0003, "grad_norm": 0.15507329924319446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47776104880, "tokens_per_second": 150378.43704593997, "data_wait_s": 1.3608001609100029e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5727ea27c6522651fa56fc722e9aa1b68242dafec68eb5f972a6675ae64a449b
-size 241065
+oid sha256:0ed7c66044e19a69289f7d7419cd14364bb16afdb01eada0c3753b58bffe7cc4
+size 243183