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": 76856.16331914364,
-      "required_pages_per_second_upper_estimate": 0.005317087709286457
+      "required_token_positions_per_second": 76919.86551138276,
+      "required_pages_per_second_upper_estimate": 0.005321494774755009
     },
     {
       "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": 9607.020414892955,
-      "required_pages_per_second_upper_estimate": 0.004022258830316453
+      "required_token_positions_per_second": 9614.983188922844,
+      "required_pages_per_second_upper_estimate": 0.004025592677521137
     },
     {
       "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": 19214.04082978591,
-      "required_pages_per_second_upper_estimate": 0.0040017779965130645
+      "required_token_positions_per_second": 19229.96637784569,
+      "required_pages_per_second_upper_estimate": 0.004005094868188971
     },
     {
       "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": 28821.061244678865,
-      "required_pages_per_second_upper_estimate": 0.0017277786144060776
+      "required_token_positions_per_second": 28844.94956676853,
+      "required_pages_per_second_upper_estimate": 0.0017292106828399971
     },
     {
       "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": 28821.061244678865,
-      "required_pages_per_second_upper_estimate": 0.0022407107413228806
+      "required_token_positions_per_second": 28844.94956676853,
+      "required_pages_per_second_upper_estimate": 0.0022425679532917277
     },
     {
       "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": 28821.061244678865,
-      "required_pages_per_second_upper_estimate": 0.3696856280021917
+      "required_token_positions_per_second": 28844.94956676853,
+      "required_pages_per_second_upper_estimate": 0.369992041748676
     }
   ],
   "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": 0.139403494710848,
-    "producer_p95_seconds": 0.31800975800069864,
-    "gpu_update_seconds": 0.9592984715337142,
-    "producer_batches_per_second": 7.1734213125302855,
-    "gpu_batches_per_second": 1.0424284304354334,
-    "producer_headroom": 6.881452100777673,
+    "producer_batch_seconds": 0.10410292924240139,
+    "producer_p95_seconds": 0.3002076049997413,
+    "gpu_update_seconds": 0.9585040159630751,
+    "producer_batches_per_second": 9.605877637424802,
+    "gpu_batches_per_second": 1.0432924467147182,
+    "producer_headroom": 9.20727229232157,
     "sustainable": true,
-    "buffer_seconds": 60.43580370662399,
+    "buffer_seconds": 60.385753005673735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384452.0738166589,
-    "upload_bytes_per_second": 4823484.397762529,
+    "generation_bytes_per_second": 383078.8298286471,
+    "upload_bytes_per_second": 4822148.389095886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.49446545481516,
+    "estimated_full_shards_per_hour": 82.19979926247177,
     "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": 1.4007872010006395,
-      "bytes": 2588923,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.1760680010011129,
       "bytes": 5785195,
@@ -562,6 +556,12 @@
       "bytes": 5814050,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.4009948579987395,
+      "bytes": 2472639,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -285,3 +285,4 @@
 {"step": 788500, "loss": 0.4945177435874939, "lr": 3e-05, "grad_norm": 0.25446166740222204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115413964021, "tokens_per_second": 151256.45725476622, "data_wait_s": 1.3527998817153275e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788600, "loss": 0.535647988319397, "lr": 3e-05, "grad_norm": 0.2629120771255029, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115428466353, "tokens_per_second": 149935.70686546213, "data_wait_s": 1.836400042520836e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788700, "loss": 0.530183732509613, "lr": 3e-05, "grad_norm": 0.2660127074293508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115443044566, "tokens_per_second": 151238.5717173458, "data_wait_s": 1.8304999684914947e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 788800, "loss": 0.5539882779121399, "lr": 3e-05, "grad_norm": 0.26287225374246087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115457688429, "tokens_per_second": 143153.21057854767, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:733c280fe360d06979c857a4d1f331adfd6cd9d7f7ee5a8413547bb1c178bbac
-size 611946
+oid sha256:eb29ffad3e7966ffeaa5712941dfe7a05f0e6e49cd19ceee002565f898da6502
+size 614064