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": 75209.78901266165,
-      "required_pages_per_second_upper_estimate": 0.006518993894149969
+      "required_token_positions_per_second": 74975.50887671053,
+      "required_pages_per_second_upper_estimate": 0.006498687085743302
     },
     {
       "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": 9401.223626582707,
-      "required_pages_per_second_upper_estimate": 0.0037502886654630233
+      "required_token_positions_per_second": 9371.938609588817,
+      "required_pages_per_second_upper_estimate": 0.003738606434334138
     },
     {
       "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": 18802.447253165414,
-      "required_pages_per_second_upper_estimate": 0.004349615260250203
+      "required_token_positions_per_second": 18743.877219177633,
+      "required_pages_per_second_upper_estimate": 0.004336066113684522
     },
     {
       "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": 28203.670879748122,
-      "required_pages_per_second_upper_estimate": 0.001717455583351528
+      "required_token_positions_per_second": 28115.815828766452,
+      "required_pages_per_second_upper_estimate": 0.0017121056716865736
     },
     {
       "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": 28203.670879748122,
-      "required_pages_per_second_upper_estimate": 0.0021711966570808843
+      "required_token_positions_per_second": 28115.815828766452,
+      "required_pages_per_second_upper_estimate": 0.002164433332057969
     },
     {
       "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": 28203.670879748122,
-      "required_pages_per_second_upper_estimate": 0.3501952007120718
+      "required_token_positions_per_second": 28115.815828766452,
+      "required_pages_per_second_upper_estimate": 0.3491043350108205
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11321958277355293,
-    "producer_p95_seconds": 0.21763889100111555,
-    "gpu_update_seconds": 0.9802979235533795,
-    "producer_batches_per_second": 8.832394321749709,
-    "gpu_batches_per_second": 1.0200980497594083,
-    "producer_headroom": 8.6583778136159,
+    "producer_batch_seconds": 0.08897254532826082,
+    "producer_p95_seconds": 0.24267770299775293,
+    "gpu_update_seconds": 0.9833611149107112,
+    "producer_batches_per_second": 11.239422187041384,
+    "gpu_batches_per_second": 1.0169204220473977,
+    "producer_headroom": 11.052410732801201,
     "sustainable": true,
-    "buffer_seconds": 60.77847126030953,
+    "buffer_seconds": 60.968389124464096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354937.2129846655,
-    "upload_bytes_per_second": 5182017.739479823,
+    "generation_bytes_per_second": 377694.770769726,
+    "upload_bytes_per_second": 5155692.176234403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.16126339106535,
+    "estimated_full_shards_per_hour": 81.0445055229076,
     "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": 0.9643511739996029,
-      "bytes": 6918272,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0286668139997346,
       "bytes": 9530498,
@@ -562,6 +556,12 @@
       "bytes": 2289232,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.238607878000039,
+      "bytes": 6177061,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -163,3 +163,4 @@
 {"step": 171300, "loss": 0.638681560754776, "lr": 0.0003, "grad_norm": 0.14773664653503873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25071024835, "tokens_per_second": 148637.95263204337, "data_wait_s": 1.2506003258749843e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 171400, "loss": 0.6007154155522585, "lr": 0.0003, "grad_norm": 0.14474890218695294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25085574445, "tokens_per_second": 148165.79361363038, "data_wait_s": 9.561001206748188e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 171500, "loss": 0.585637129843235, "lr": 0.0003, "grad_norm": 0.14237854064749242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25100065919, "tokens_per_second": 148326.70183672974, "data_wait_s": 1.4298999303719029e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 171600, "loss": 0.5917437970638275, "lr": 0.0003, "grad_norm": 0.1492916433478241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25114568192, "tokens_per_second": 148368.72933766892, "data_wait_s": 1.4881999959470704e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c59e7bfbf376a7aab8ee713fe0bf97e271e15cf22bc1d016b422ec5de7ba27e
-size 351984
+oid sha256:27ad561b1c0f57831fbc1246ae03f84ab9b2b8991f19d9a4b2710274b4b23a54
+size 354102