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": 74164.61332669585,
-      "required_pages_per_second_upper_estimate": 0.006428400714664011
+      "required_token_positions_per_second": 74159.67083950838,
+      "required_pages_per_second_upper_estimate": 0.006427972312401221
     },
     {
       "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": 9270.576665836981,
-      "required_pages_per_second_upper_estimate": 0.0036981716394754193
+      "required_token_positions_per_second": 9269.958854938548,
+      "required_pages_per_second_upper_estimate": 0.0036979251854709384
     },
     {
       "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": 18541.153331673962,
-      "required_pages_per_second_upper_estimate": 0.0042891695101291115
+      "required_token_positions_per_second": 18539.917709877096,
+      "required_pages_per_second_upper_estimate": 0.004288883670772608
     },
     {
       "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": 27811.72999751094,
-      "required_pages_per_second_upper_estimate": 0.0016935884399781693
+      "required_token_positions_per_second": 27809.876564815644,
+      "required_pages_per_second_upper_estimate": 0.0016934755756512355
     },
     {
       "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": 27811.72999751094,
-      "required_pages_per_second_upper_estimate": 0.0021410239630044636
+      "required_token_positions_per_second": 27809.876564815644,
+      "required_pages_per_second_upper_estimate": 0.0021408812806249494
     },
     {
       "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": 27811.72999751094,
-      "required_pages_per_second_upper_estimate": 0.34532860669643695
+      "required_token_positions_per_second": 27809.876564815644,
+      "required_pages_per_second_upper_estimate": 0.34530559326540156
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09720908353125779,
-    "producer_p95_seconds": 0.16321426400008932,
-    "gpu_update_seconds": 0.9941129157544103,
-    "producer_batches_per_second": 10.28710449346483,
-    "gpu_batches_per_second": 1.0059219472479362,
-    "producer_headroom": 10.226543442668618,
+    "producer_batch_seconds": 0.0707124810234383,
+    "producer_p95_seconds": 0.16263690300002054,
+    "gpu_update_seconds": 0.9941791699636507,
+    "producer_batches_per_second": 14.141775051967711,
+    "gpu_batches_per_second": 1.005854910475103,
+    "producer_headroom": 14.059458182977924,
     "sustainable": true,
-    "buffer_seconds": 60.64088786101903,
+    "buffer_seconds": 60.644929367782694,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 529933.2179946193,
-    "upload_bytes_per_second": 5555791.942111689,
+    "generation_bytes_per_second": 641109.0709567142,
+    "upload_bytes_per_second": 5600949.953575108,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.71133236769613,
+    "estimated_full_shards_per_hour": 68.78354118598017,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 1054926,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.207006858999648,
+      "bytes": 7769728,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 84200, "loss": 0.5921671111136675, "lr": 0.0003, "grad_norm": 0.1334533288816832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12322412062, "tokens_per_second": 148253.383593555, "data_wait_s": 1.087200007532374e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84300, "loss": 0.650002021342516, "lr": 0.0003, "grad_norm": 0.16830847416344066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12336900344, "tokens_per_second": 146661.90606472787, "data_wait_s": 1.3800000033370452e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84400, "loss": 0.6230411995202303, "lr": 0.0003, "grad_norm": 0.1334623499219814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12351402779, "tokens_per_second": 146855.37327678414, "data_wait_s": 1.076500029739691e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 84500, "loss": 0.6755691282451153, "lr": 0.0003, "grad_norm": 0.15844106523682092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12365969201, "tokens_per_second": 147706.36542072735, "data_wait_s": 1.2414000138960546e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1d738a2848f33dde8b61a470ac1247452f666e35aab5d602693b39279feb5ce
-size 53115
+oid sha256:48832adb1ffcb68fce073fb0590e54e5a5f410859f574930408dc2e9dc357d57
+size 55233