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": 76542.64076637056,
-      "required_pages_per_second_upper_estimate": 0.0052953974916130665
+      "required_token_positions_per_second": 76362.0519283529,
+      "required_pages_per_second_upper_estimate": 0.00528290393677517
     },
     {
       "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": 9567.83009579632,
-      "required_pages_per_second_upper_estimate": 0.004005850662097616
+      "required_token_positions_per_second": 9545.256491044112,
+      "required_pages_per_second_upper_estimate": 0.003996399565178338
     },
     {
       "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": 19135.66019159264,
-      "required_pages_per_second_upper_estimate": 0.003985453376613837
+      "required_token_positions_per_second": 19090.512982088225,
+      "required_pages_per_second_upper_estimate": 0.003976050403486048
     },
     {
       "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": 28703.490287388962,
-      "required_pages_per_second_upper_estimate": 0.0017207304150370052
+      "required_token_positions_per_second": 28635.769473132335,
+      "required_pages_per_second_upper_estimate": 0.0017166706556260164
     },
     {
       "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": 28703.490287388962,
-      "required_pages_per_second_upper_estimate": 0.002231570116533579
+      "required_token_positions_per_second": 28635.769473132335,
+      "required_pages_per_second_upper_estimate": 0.002226305121097508
     },
     {
       "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": 28703.490287388962,
-      "required_pages_per_second_upper_estimate": 0.3681775539999354
+      "required_token_positions_per_second": 28635.769473132335,
+      "required_pages_per_second_upper_estimate": 0.36730890410759653
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12105641596096461,
-    "producer_p95_seconds": 0.26598942099735723,
-    "gpu_update_seconds": 0.9632278069035842,
-    "producer_batches_per_second": 8.260611319621887,
-    "gpu_batches_per_second": 1.0381760086584548,
-    "producer_headroom": 7.956850525082313,
+    "producer_batch_seconds": 0.0663623994844329,
+    "producer_p95_seconds": 0.23299714799941285,
+    "gpu_update_seconds": 0.9655057471370163,
+    "producer_batches_per_second": 15.06877400107537,
+    "gpu_batches_per_second": 1.0357266157816962,
+    "producer_headroom": 14.548987900347122,
     "sustainable": true,
-    "buffer_seconds": 60.68335183492581,
+    "buffer_seconds": 60.82686206963203,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382870.5919952992,
-    "upload_bytes_per_second": 4779082.206989607,
+    "generation_bytes_per_second": 374086.07476035604,
+    "upload_bytes_per_second": 4795324.289960057,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.15511627096397,
+    "estimated_full_shards_per_hour": 80.27016336544047,
     "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": 1.1759212130000378,
-      "bytes": 6164782,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.4110855310009356,
       "bytes": 8184458,
@@ -562,6 +556,12 @@
       "bytes": 1175820,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.1064946840015182,
+      "bytes": 7293995,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 947500, "loss": 0.5070681571960449, "lr": 3e-05, "grad_norm": 0.35016341403940265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138687659586, "tokens_per_second": 153170.24428802286, "data_wait_s": 1.383500057272613e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 947600, "loss": 0.5208671689033508, "lr": 3e-05, "grad_norm": 0.3406378189495496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138702328071, "tokens_per_second": 152857.38962968843, "data_wait_s": 7.000999175943434e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 947700, "loss": 0.5153999924659729, "lr": 3e-05, "grad_norm": 0.3330193206786876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138717057800, "tokens_per_second": 153810.06986681506, "data_wait_s": 1.4757999451830983e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 947800, "loss": 0.5304925441741943, "lr": 3e-05, "grad_norm": 0.3326355692209255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138731718809, "tokens_per_second": 152488.53888807967, "data_wait_s": 1.5317000361392274e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ec32929c11f5f8369e35726e6518783608ee650e30ae713f85984ffdd9ca779
-size 484083
+oid sha256:7c2468f424e63c2e1cee8d1edad9ea572838c03d1ff90df6b0498f21c49a7734
+size 486201