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": 76286.95893242754,
-      "required_pages_per_second_upper_estimate": 0.005277708829077298
+      "required_token_positions_per_second": 77402.13921505402,
+      "required_pages_per_second_upper_estimate": 0.0053548595886041515
     },
     {
       "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": 9535.869866553443,
-      "required_pages_per_second_upper_estimate": 0.003992469581519103
+      "required_token_positions_per_second": 9675.267401881752,
+      "required_pages_per_second_upper_estimate": 0.004050832418609513
     },
     {
       "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": 19071.739733106886,
-      "required_pages_per_second_upper_estimate": 0.003972140430807103
+      "required_token_positions_per_second": 19350.534803763505,
+      "required_pages_per_second_upper_estimate": 0.0040302060917044415
     },
     {
       "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": 28607.60959966033,
-      "required_pages_per_second_upper_estimate": 0.0017149825142090068
+      "required_token_positions_per_second": 29025.802205645254,
+      "required_pages_per_second_upper_estimate": 0.0017400525223946665
     },
     {
       "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": 28607.60959966033,
-      "required_pages_per_second_upper_estimate": 0.0022241158147972527
+      "required_token_positions_per_second": 29025.802205645254,
+      "required_pages_per_second_upper_estimate": 0.0022566284504777043
     },
     {
       "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": 28607.60959966033,
-      "required_pages_per_second_upper_estimate": 0.3669476994864141
+      "required_token_positions_per_second": 29025.802205645254,
+      "required_pages_per_second_upper_estimate": 0.3723118252157522
     }
   ],
   "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": 0.07464658006253444,
-    "producer_p95_seconds": 0.19409570499919937,
-    "gpu_update_seconds": 0.9664561418066988,
-    "producer_batches_per_second": 13.396461018874003,
-    "gpu_batches_per_second": 1.0347081018395663,
-    "producer_headroom": 12.947092030164807,
+    "producer_batch_seconds": 0.10974192438277441,
+    "producer_p95_seconds": 0.19312320700009877,
+    "gpu_update_seconds": 0.9525318130440066,
+    "producer_batches_per_second": 9.112287811830685,
+    "gpu_batches_per_second": 1.0498337024611275,
+    "producer_headroom": 8.679744030381887,
     "sustainable": true,
-    "buffer_seconds": 60.88673693382202,
+    "buffer_seconds": 60.00950422177242,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343068.0565775419,
-    "upload_bytes_per_second": 5918343.718301311,
+    "generation_bytes_per_second": 329672.39272049506,
+    "upload_bytes_per_second": 5845001.442711912,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.61441872591679,
+    "estimated_full_shards_per_hour": 70.74002109728946,
     "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": 1.4391079690003608,
-      "bytes": 12352688,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9268819079998138,
       "bytes": 1009442,
@@ -562,6 +556,12 @@
       "bytes": 6495644,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2645150809985353,
+      "bytes": 6000960,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -84,3 +84,4 @@
 {"step": 493400, "loss": 0.599118709564209, "lr": 0.0003, "grad_norm": 0.15573941228074265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72217316920, "tokens_per_second": 154023.2147727918, "data_wait_s": 1.1088000974268652e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 493500, "loss": 0.5740088820457458, "lr": 0.0003, "grad_norm": 0.15058203666860437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72232061963, "tokens_per_second": 153394.80132307735, "data_wait_s": 1.3472999853547662e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 493600, "loss": 0.5430557131767273, "lr": 0.0003, "grad_norm": 0.15844527716267476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72246786544, "tokens_per_second": 153937.86656768378, "data_wait_s": 1.688900010776706e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 493700, "loss": 0.5841078162193298, "lr": 0.0003, "grad_norm": 0.16528281394083058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72261458633, "tokens_per_second": 153526.9971127577, "data_wait_s": 1.042300027620513e-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:db97e7b84ef0ebff51aafd463d936bda01f43d9465c270dd435c9e3511a6c069
-size 183096
+oid sha256:22782a594638a0a85b49950d9f01c0acb80322b1021ed549474e04f48895109b
+size 185214