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": 74608.41523468573,
-      "required_pages_per_second_upper_estimate": 0.006466868339242922
+      "required_token_positions_per_second": 74013.55310427283,
+      "required_pages_per_second_upper_estimate": 0.006415307197443017
     },
     {
       "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": 9326.051904335716,
-      "required_pages_per_second_upper_estimate": 0.003720301541541294
+      "required_token_positions_per_second": 9251.694138034103,
+      "required_pages_per_second_upper_estimate": 0.003690639116815902
     },
     {
       "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": 18652.103808671432,
-      "required_pages_per_second_upper_estimate": 0.004314835950320773
+      "required_token_positions_per_second": 18503.388276068206,
+      "required_pages_per_second_upper_estimate": 0.004280433229156998
     },
     {
       "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": 27978.15571300715,
-      "required_pages_per_second_upper_estimate": 0.0017037228928836411
+      "required_token_positions_per_second": 27755.08241410231,
+      "required_pages_per_second_upper_estimate": 0.0016901388993554845
     },
     {
       "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": 27978.15571300715,
-      "required_pages_per_second_upper_estimate": 0.0021538358752792278
+      "required_token_positions_per_second": 27755.08241410231,
+      "required_pages_per_second_upper_estimate": 0.00213666307522312
     },
     {
       "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": 27978.15571300715,
-      "required_pages_per_second_upper_estimate": 0.34739505709186025
+      "required_token_positions_per_second": 27755.08241410231,
+      "required_pages_per_second_upper_estimate": 0.3446252332977676
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08643227771881357,
-    "producer_p95_seconds": 0.24020139500134974,
-    "gpu_update_seconds": 0.9881995183530393,
-    "producer_batches_per_second": 11.569751791723657,
-    "gpu_batches_per_second": 1.011941395869761,
-    "producer_headroom": 11.433223148045531,
+    "producer_batch_seconds": 0.4174442869764903,
+    "producer_p95_seconds": 0.30998019899925566,
+    "gpu_update_seconds": 0.9961418808813227,
+    "producer_batches_per_second": 2.395529250724464,
+    "gpu_batches_per_second": 1.0038730618526588,
+    "producer_headroom": 2.3862870135228937,
     "sustainable": true,
-    "buffer_seconds": 60.2801706195354,
+    "buffer_seconds": 60.764654733760686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 253098.74297243578,
-    "upload_bytes_per_second": 5872959.067682285,
+    "generation_bytes_per_second": 258579.42582483374,
+    "upload_bytes_per_second": 5899148.882159589,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.309098404691746,
+    "estimated_full_shards_per_hour": 55.48512536104926,
     "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": 1.1277135729997099,
-      "bytes": 5536865,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 0.9057988900003693,
       "bytes": 5888539,
@@ -562,6 +556,12 @@
       "bytes": 1129396,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.0460155590008071,
+      "bytes": 6970566,
+      "pages": 24,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -192,3 +192,4 @@
 {"step": 19200, "loss": 0.7024876587092876, "lr": 0.0003, "grad_norm": 0.16395474842740984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2809542028, "tokens_per_second": 147541.79972931175, "data_wait_s": 1.269100175704807e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 19300, "loss": 0.7050697933882475, "lr": 0.0003, "grad_norm": 0.1251860720431679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2824197435, "tokens_per_second": 149184.74466385331, "data_wait_s": 1.223800063598901e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 19400, "loss": 0.7558563780039549, "lr": 0.0003, "grad_norm": 0.1484141513677437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2838716086, "tokens_per_second": 146863.99147552872, "data_wait_s": 9.840998245636001e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 19500, "loss": 0.6772917918860912, "lr": 0.0003, "grad_norm": 0.136990950203081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2853181602, "tokens_per_second": 147552.31416240722, "data_wait_s": 1.0497999028302729e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a6051d83509d55048628aa255f2e3b5a01be6b708ba2ab8aec6862d022254c6
-size 407632
+oid sha256:bfad78b6dcb9497354c4d436771df70ce7768e97247aa2f0aa6374abb86fd2c1
+size 409750