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": 77143.9981967053,
-      "required_pages_per_second_upper_estimate": 0.005337000793985102
+      "required_token_positions_per_second": 76577.43011913888,
+      "required_pages_per_second_upper_estimate": 0.005297804299759999
     },
     {
       "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": 9642.999774588163,
-      "required_pages_per_second_upper_estimate": 0.004037322636886369
+      "required_token_positions_per_second": 9572.17876489236,
+      "required_pages_per_second_upper_estimate": 0.00400767135903759
     },
     {
       "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": 19285.999549176326,
-      "required_pages_per_second_upper_estimate": 0.004016765100083044
+      "required_token_positions_per_second": 19144.35752978472,
+      "required_pages_per_second_upper_estimate": 0.003987264802795015
     },
     {
       "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": 28928.99932376449,
-      "required_pages_per_second_upper_estimate": 0.0017342493374353568
+      "required_token_positions_per_second": 28716.536294677077,
+      "required_pages_per_second_upper_estimate": 0.0017215125032538278
     },
     {
       "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": 28928.99932376449,
-      "required_pages_per_second_upper_estimate": 0.002249102452202353
+      "required_token_positions_per_second": 28716.536294677077,
+      "required_pages_per_second_upper_estimate": 0.002232584386216908
     },
     {
       "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": 28928.99932376449,
-      "required_pages_per_second_upper_estimate": 0.37107014178582226
+      "required_token_positions_per_second": 28716.536294677077,
+      "required_pages_per_second_upper_estimate": 0.368344894173716
     }
   ],
   "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": 0.0857974322578059,
-    "producer_p95_seconds": 0.175905332000184,
-    "gpu_update_seconds": 0.9557191968713492,
-    "producer_batches_per_second": 11.655360465744236,
-    "gpu_batches_per_second": 1.046332440819028,
-    "producer_headroom": 11.139251743567156,
+    "producer_batch_seconds": 0.491669049414142,
+    "producer_p95_seconds": 0.2010451969999849,
+    "gpu_update_seconds": 0.9627902096648354,
+    "producer_batches_per_second": 2.033888448320206,
+    "gpu_batches_per_second": 1.038647869454466,
+    "producer_headroom": 1.958207885593098,
     "sustainable": true,
-    "buffer_seconds": 60.210309402895,
+    "buffer_seconds": 60.65578320888463,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304097.3337023473,
-    "upload_bytes_per_second": 4496015.548259048,
+    "generation_bytes_per_second": 292133.8636338746,
+    "upload_bytes_per_second": 4497947.86412636,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.2522087888986,
+    "estimated_full_shards_per_hour": 62.685126607534194,
     "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": 0.8292768119999891,
-      "bytes": 2252671,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.5412007810000432,
       "bytes": 6087939,
@@ -562,6 +556,12 @@
       "bytes": 6121395,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8000806340005511,
+      "bytes": 2307263,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -84,3 +84,4 @@
 {"step": 448600, "loss": 0.5526136159896851, "lr": 0.0003, "grad_norm": 0.18028248759067692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65661268365, "tokens_per_second": 155077.3128269652, "data_wait_s": 1.3885999578633346e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448700, "loss": 0.5940396189689636, "lr": 0.0003, "grad_norm": 0.15803821648863708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65676020290, "tokens_per_second": 152803.27240567782, "data_wait_s": 1.5137999071157537e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448800, "loss": 0.5863661766052246, "lr": 0.0003, "grad_norm": 0.1476474418774333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65690699324, "tokens_per_second": 152705.65303955795, "data_wait_s": 1.2272999811102636e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 448900, "loss": 0.6093892455101013, "lr": 0.0003, "grad_norm": 0.1593004657932762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65705297294, "tokens_per_second": 152426.27186727803, "data_wait_s": 1.1653999536065385e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c9a808e7c057b1ffabb08ed2df0cd9a85923567adc93c231b3c8694c51674faf
-size 183096
+oid sha256:4214a951d8bb152340d75c85ff91cb8f19c7584053ce093df782552999caf8d0
+size 185214