Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76296.39665152293,
-      "required_pages_per_second_upper_estimate": 0.0052783617523828414
+      "required_token_positions_per_second": 76761.01615405783,
+      "required_pages_per_second_upper_estimate": 0.00531050520763398
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9537.049581440366,
-      "required_pages_per_second_upper_estimate": 0.003992963503507009
+      "required_token_positions_per_second": 9595.127019257228,
+      "required_pages_per_second_upper_estimate": 0.004017279313926116
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19074.09916288073,
-      "required_pages_per_second_upper_estimate": 0.003972631837806648
+      "required_token_positions_per_second": 19190.254038514457,
+      "required_pages_per_second_upper_estimate": 0.00399682383519109
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28611.148744321094,
-      "required_pages_per_second_upper_estimate": 0.0017151946805274586
+      "required_token_positions_per_second": 28785.381057771687,
+      "required_pages_per_second_upper_estimate": 0.001725639641681483
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28611.148744321094,
-      "required_pages_per_second_upper_estimate": 0.002224390967727575
+      "required_token_positions_per_second": 28785.381057771687,
+      "required_pages_per_second_upper_estimate": 0.002237936763731404
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28611.148744321094,
-      "required_pages_per_second_upper_estimate": 0.3669930958340849
+      "required_token_positions_per_second": 28785.381057771687,
+      "required_pages_per_second_upper_estimate": 0.3692279608749463
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.9118546186877552,
-    "producer_p95_seconds": 0.4556006439961493,
-    "gpu_update_seconds": 0.9663365930208494,
-    "producer_batches_per_second": 1.0966660468738914,
-    "gpu_batches_per_second": 1.0348361090972618,
-    "producer_headroom": 1.0597485314177593,
+    "producer_batch_seconds": 0.08692735840600108,
+    "producer_p95_seconds": 0.3136482830013847,
+    "gpu_update_seconds": 0.9604875455534535,
+    "producer_batches_per_second": 11.503858144744504,
+    "gpu_batches_per_second": 1.0411379144159318,
+    "producer_headroom": 11.049312473840752,
     "sustainable": true,
-    "buffer_seconds": 60.87920536031351,
+    "buffer_seconds": 60.51071536986757,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 254666.2865500591,
-    "upload_bytes_per_second": 5330419.314515897,
+    "generation_bytes_per_second": 254795.84893678385,
+    "upload_bytes_per_second": 5438731.173210582,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.6454567659028,
+    "estimated_full_shards_per_hour": 54.67325783803593,
     "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": 1.0647863270023663,
-      "bytes": 1121901,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.296304608000355,
       "bytes": 7362354,
@@ -562,6 +556,12 @@
       "bytes": 8148657,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2602131539970287,
+      "bytes": 10952650,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 523100, "loss": 0.5512901544570923, "lr": 0.0003, "grad_norm": 0.18323662469231008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76563821788, "tokens_per_second": 151782.80051866503, "data_wait_s": 9.128001693170518e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 523200, "loss": 0.5553374886512756, "lr": 0.0003, "grad_norm": 0.17085884588890812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76578318938, "tokens_per_second": 141690.74500809534, "data_wait_s": 8.859002264216542e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 523300, "loss": 0.5911731123924255, "lr": 0.0003, "grad_norm": 0.1815925787559326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76593048711, "tokens_per_second": 154341.8191260696, "data_wait_s": 1.0716001270338893e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 523400, "loss": 0.5877964496612549, "lr": 0.0003, "grad_norm": 0.18020763398318995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76607715438, "tokens_per_second": 153205.58404012618, "data_wait_s": 1.5669000276830047e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a5abeedb667fd556fb5e4c99b226b5222a96c676149c1067f2e1881a1295a66
-size 816840
+oid sha256:4f5b602bd5a5d45699f7e70d839f12d326f847ef6975755314163abbc267d59e
+size 818958