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": 77146.38679104972,
-      "required_pages_per_second_upper_estimate": 0.005337166042484152
+      "required_token_positions_per_second": 76863.71216586295,
+      "required_pages_per_second_upper_estimate": 0.005317609956018236
     },
     {
       "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": 9643.298348881215,
-      "required_pages_per_second_upper_estimate": 0.004037447643707929
+      "required_token_positions_per_second": 9607.964020732868,
+      "required_pages_per_second_upper_estimate": 0.004022653898376894
     },
     {
       "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": 19286.59669776243,
-      "required_pages_per_second_upper_estimate": 0.004016889470385663
+      "required_token_positions_per_second": 19215.928041465737,
+      "required_pages_per_second_upper_estimate": 0.004002171052936853
     },
     {
       "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": 28929.895046643647,
-      "required_pages_per_second_upper_estimate": 0.0017343030346542727
+      "required_token_positions_per_second": 28823.892062198604,
+      "required_pages_per_second_upper_estimate": 0.0017279483175939778
     },
     {
       "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": 28929.895046643647,
-      "required_pages_per_second_upper_estimate": 0.002249172090716339
+      "required_token_positions_per_second": 28823.892062198604,
+      "required_pages_per_second_upper_estimate": 0.00224093082493359
     },
     {
       "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": 28929.895046643647,
-      "required_pages_per_second_upper_estimate": 0.3710816311571638
+      "required_token_positions_per_second": 28823.892062198604,
+      "required_pages_per_second_upper_estimate": 0.3697219386898398
     }
   ],
   "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": 0.3347509299219098,
-    "producer_p95_seconds": 0.2531154989992501,
-    "gpu_update_seconds": 0.9556896060432695,
-    "producer_batches_per_second": 2.9872956595916804,
-    "gpu_batches_per_second": 1.0463648382032569,
-    "producer_headroom": 2.854927412049942,
+    "producer_batch_seconds": 0.11312406903925876,
+    "producer_p95_seconds": 0.23695643300015945,
+    "gpu_update_seconds": 0.9592042580626806,
+    "producer_batches_per_second": 8.83985175297185,
+    "gpu_batches_per_second": 1.0425308182218824,
+    "producer_headroom": 8.47922344209345,
     "sustainable": true,
-    "buffer_seconds": 60.208445180725974,
+    "buffer_seconds": 60.429868257948876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357318.37698609236,
-    "upload_bytes_per_second": 1902842.841712294,
+    "generation_bytes_per_second": 356622.7802336352,
+    "upload_bytes_per_second": 1939755.3517491233,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.67220575511054,
+    "estimated_full_shards_per_hour": 76.52294688469689,
     "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": 1.0003182039999956,
-      "bytes": 2721093,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0431637549991137,
       "bytes": 11688824,
@@ -562,6 +556,12 @@
       "bytes": 6030424,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2485899509993033,
+      "bytes": 11733880,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -189,3 +189,4 @@
 {"step": 663900, "loss": 0.5550321340560913, "lr": 0.0001, "grad_norm": 0.2184161685445405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97178131334, "tokens_per_second": 154107.05724833082, "data_wait_s": 1.5232999430736527e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 664000, "loss": 0.5500065088272095, "lr": 0.0001, "grad_norm": 0.1818294185679898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97192694370, "tokens_per_second": 151407.37875300326, "data_wait_s": 9.307001164415851e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 664100, "loss": 0.5199372172355652, "lr": 0.0001, "grad_norm": 0.17990135708849617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97207173670, "tokens_per_second": 150201.06743699362, "data_wait_s": 1.4229000953491777e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 664200, "loss": 0.5283790230751038, "lr": 0.0001, "grad_norm": 0.19205047311701712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97221728887, "tokens_per_second": 151822.64087532778, "data_wait_s": 1.0331001249141991e-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 sha256:722fa94bc74deb478416668f5f1193d3affe254a1d927f209ad790f809511b30
-size 407052
+oid sha256:a2ae5bbd275e0852aeb95bdbe0fc74adc3eadb51d8d0e06eb62c25b340ce3c88
+size 409170