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": 74751.85584934599,
-      "required_pages_per_second_upper_estimate": 0.0064793014081747586
+      "required_token_positions_per_second": 74549.9611970052,
+      "required_pages_per_second_upper_estimate": 0.006461801691407178
     },
     {
       "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": 9343.981981168248,
-      "required_pages_per_second_upper_estimate": 0.0037274541172683294
+      "required_token_positions_per_second": 9318.74514962565,
+      "required_pages_per_second_upper_estimate": 0.0037173867678417308
     },
     {
       "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": 18687.963962336496,
-      "required_pages_per_second_upper_estimate": 0.0043231315657540365
+      "required_token_positions_per_second": 18637.4902992513,
+      "required_pages_per_second_upper_estimate": 0.004311455372105408
     },
     {
       "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": 28031.945943504743,
-      "required_pages_per_second_upper_estimate": 0.001706998435705416
+      "required_token_positions_per_second": 27956.23544887695,
+      "required_pages_per_second_upper_estimate": 0.0017023880638048496
     },
     {
       "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": 28031.945943504743,
-      "required_pages_per_second_upper_estimate": 0.0021579767961238208
+      "required_token_positions_per_second": 27956.23544887695,
+      "required_pages_per_second_upper_estimate": 0.0021521483926673106
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28031.945943504743,
-      "required_pages_per_second_upper_estimate": 0.34806295173031954
+      "required_token_positions_per_second": 27956.23544887695,
+      "required_pages_per_second_upper_estimate": 0.34712288077376796
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10625565530438053,
-    "producer_p95_seconds": 0.2735050070004945,
-    "gpu_update_seconds": 0.9863032718356927,
-    "producer_batches_per_second": 9.411263778246857,
-    "gpu_batches_per_second": 1.013886933720513,
-    "producer_headroom": 9.282360256593618,
+    "producer_batch_seconds": 0.09850960999983727,
+    "producer_p95_seconds": 0.3052236350013118,
+    "gpu_update_seconds": 0.9889743578157862,
+    "producer_batches_per_second": 10.151293868706333,
+    "gpu_batches_per_second": 1.0111485622423666,
+    "producer_headroom": 10.039369334803173,
     "sustainable": true,
-    "buffer_seconds": 60.16449958197725,
+    "buffer_seconds": 60.32743582676296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317105.94055001036,
-    "upload_bytes_per_second": 4409809.874766142,
+    "generation_bytes_per_second": 318578.5408104754,
+    "upload_bytes_per_second": 4467586.492257718,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.04355299353821,
+    "estimated_full_shards_per_hour": 68.35953872905442,
     "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": 1.1722490869979083,
-      "bytes": 806757,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 0.9350021480022406,
       "bytes": 6160019,
@@ -562,6 +556,12 @@
       "bytes": 5907251,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.345177949002391,
+      "bytes": 6941210,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -259,3 +259,4 @@
 {"step": 263000, "loss": 0.6872336938977242, "lr": 0.0003, "grad_norm": 0.17677581162894507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38495777649, "tokens_per_second": 149705.37010786976, "data_wait_s": 1.454499943065457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 263100, "loss": 0.5863943379372358, "lr": 0.0003, "grad_norm": 0.1542861177462755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38510485407, "tokens_per_second": 148934.96212251476, "data_wait_s": 1.3859000318916515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 263200, "loss": 0.5569163095206022, "lr": 0.0003, "grad_norm": 0.1477946999970373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38525157234, "tokens_per_second": 149471.69233585094, "data_wait_s": 1.0715000826166943e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 263300, "loss": 0.6413752101361752, "lr": 0.0003, "grad_norm": 0.16248475905404894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38539889874, "tokens_per_second": 149717.19177236749, "data_wait_s": 1.1845997505588457e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b3cee6ad8258366e68e8950cb54c0703324bb7621e56adb3218db4357fbd5d5
-size 556878
+oid sha256:a166d38ea58b0553f2ece60fcc216f12d8af3cabb2a4193c229fbb6774e9cfb8
+size 558996