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": 77081.95241618187,
-      "required_pages_per_second_upper_estimate": 0.005332708322922446
+      "required_token_positions_per_second": 77388.53991736824,
+      "required_pages_per_second_upper_estimate": 0.005353918757635531
     },
     {
       "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": 9635.244052022734,
-      "required_pages_per_second_upper_estimate": 0.004034075477806127
+      "required_token_positions_per_second": 9673.56748967103,
+      "required_pages_per_second_upper_estimate": 0.004050120700865087
     },
     {
       "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": 19270.488104045467,
-      "required_pages_per_second_upper_estimate": 0.004013534475126603
+      "required_token_positions_per_second": 19347.13497934206,
+      "required_pages_per_second_upper_estimate": 0.004029497997936854
     },
     {
       "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": 28905.732156068203,
-      "required_pages_per_second_upper_estimate": 0.0017328545062588725
+      "required_token_positions_per_second": 29020.702469013086,
+      "required_pages_per_second_upper_estimate": 0.0017397468009704151
     },
     {
       "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": 28905.732156068203,
-      "required_pages_per_second_upper_estimate": 0.002247293532255422
+      "required_token_positions_per_second": 29020.702469013086,
+      "required_pages_per_second_upper_estimate": 0.002256231968385924
     },
     {
       "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": 28905.732156068203,
-      "required_pages_per_second_upper_estimate": 0.3707716955409526
+      "required_token_positions_per_second": 29020.702469013086,
+      "required_pages_per_second_upper_estimate": 0.37224641126990526
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05034595199999359,
-    "producer_p95_seconds": 0.1634836670000368,
-    "gpu_update_seconds": 0.9564884864608363,
-    "producer_batches_per_second": 19.862570083094813,
-    "gpu_batches_per_second": 1.0454908910614946,
-    "producer_headroom": 18.998319596001643,
+    "producer_batch_seconds": 0.22007113467966377,
+    "producer_p95_seconds": 0.16821561199958524,
+    "gpu_update_seconds": 0.9526991991155696,
+    "producer_batches_per_second": 4.543985295734505,
+    "gpu_batches_per_second": 1.0496492501813182,
+    "producer_headroom": 4.329051152039187,
     "sustainable": true,
-    "buffer_seconds": 60.25877464703269,
+    "buffer_seconds": 60.02004954428088,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453459.2910779464,
-    "upload_bytes_per_second": 2009161.748861214,
+    "generation_bytes_per_second": 453601.85631606035,
+    "upload_bytes_per_second": 1969860.9920954169,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.30180787328523,
+    "estimated_full_shards_per_hour": 97.3323990546356,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 7780089,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.7348476629995275,
+      "bytes": 1091907,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 650500, "loss": 0.533645749092102, "lr": 0.0001, "grad_norm": 0.17229197103661847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95216212674, "tokens_per_second": 152768.75228491038, "data_wait_s": 8.107999747153372e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 650600, "loss": 0.570576548576355, "lr": 0.0001, "grad_norm": 0.18087243068328412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95230799015, "tokens_per_second": 152227.03190944484, "data_wait_s": 1.034399974741973e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 650700, "loss": 0.5605326294898987, "lr": 0.0001, "grad_norm": 0.20141436027587253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95245328463, "tokens_per_second": 152014.6132398883, "data_wait_s": 1.1510000149428379e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 650800, "loss": 0.5625901222229004, "lr": 0.0001, "grad_norm": 0.18249236859675003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95259860361, "tokens_per_second": 150093.4890971094, "data_wait_s": 1.2041000445606187e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a772eb69be0bb835018e64e14a6d5f7f328259e2159c2d3869f209f3d8238f7
-size 121674
+oid sha256:878c0f6653a597964bbd9d0c9155cdf6b2bc4ac1c7fc8bfdbb820d99a0602d54
+size 123792