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": 75387.55786154333,
-      "required_pages_per_second_upper_estimate": 0.006534402447420005
+      "required_token_positions_per_second": 75512.56195927144,
+      "required_pages_per_second_upper_estimate": 0.006545237485791088
     },
     {
       "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": 9423.444732692917,
-      "required_pages_per_second_upper_estimate": 0.003759152996925529
+      "required_token_positions_per_second": 9439.07024490893,
+      "required_pages_per_second_upper_estimate": 0.003765386247370723
     },
     {
       "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": 18846.889465385833,
-      "required_pages_per_second_upper_estimate": 0.004359896183891176
+      "required_token_positions_per_second": 18878.14048981786,
+      "required_pages_per_second_upper_estimate": 0.00436712555839428
     },
     {
       "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": 28270.334198078748,
-      "required_pages_per_second_upper_estimate": 0.0017215150296824623
+      "required_token_positions_per_second": 28317.21073472679,
+      "required_pages_per_second_upper_estimate": 0.0017243695648221477
     },
     {
       "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": 28270.334198078748,
-      "required_pages_per_second_upper_estimate": 0.002176328583861852
+      "required_token_positions_per_second": 28317.21073472679,
+      "required_pages_per_second_upper_estimate": 0.002179937269415577
     },
     {
       "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": 28270.334198078748,
-      "required_pages_per_second_upper_estimate": 0.35102293601796375
+      "required_token_positions_per_second": 28317.21073472679,
+      "required_pages_per_second_upper_estimate": 0.3516049857174564
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10757936447672023,
-    "producer_p95_seconds": 0.22268259500197018,
-    "gpu_update_seconds": 0.9779863161956874,
-    "producer_batches_per_second": 9.295462980880467,
-    "gpu_batches_per_second": 1.022509194085603,
-    "producer_headroom": 9.090835598004672,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.63515160413262,
+    "buffer_seconds": 60.53477569024203,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 610880.2736134968,
-    "upload_bytes_per_second": 5733726.810197932,
+    "generation_bytes_per_second": 634204.3550508835,
+    "upload_bytes_per_second": 5722526.054790809,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.54034307624663,
+    "estimated_full_shards_per_hour": 68.04274553606452,
     "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": 0.831416236998848,
-      "bytes": 5844217,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.3119218800002272,
       "bytes": 12680620,
@@ -562,6 +556,12 @@
       "bytes": 7383726,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.007008101001702,
+      "bytes": 6024431,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 371900, "loss": 0.5916075948625803, "lr": 0.0003, "grad_norm": 0.1549158219388106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54435179152, "tokens_per_second": 148884.56980442794, "data_wait_s": 1.1111998901469633e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:2bdafd7f999370ad4b9ca3fd281bce1c09a41ee416212e4b8b05f55feb591837
-size 364692
+oid sha256:80a64ac2def6958f79ae98d1905736ca4c0e9194ab6711dc44a3c10eb917f4a3
+size 366810