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": 76973.87039167676,
-      "required_pages_per_second_upper_estimate": 0.005325230957682334
+      "required_token_positions_per_second": 76984.27088791992,
+      "required_pages_per_second_upper_estimate": 0.005325950488144912
     },
     {
       "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": 9621.733798959594,
-      "required_pages_per_second_upper_estimate": 0.004028419016974757
+      "required_token_positions_per_second": 9623.03386098999,
+      "required_pages_per_second_upper_estimate": 0.004028963325798501
     },
     {
       "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": 19243.46759791919,
-      "required_pages_per_second_upper_estimate": 0.004007906816279164
+      "required_token_positions_per_second": 19246.06772197998,
+      "required_pages_per_second_upper_estimate": 0.0040084483535511445
     },
     {
       "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": 28865.201396878783,
-      "required_pages_per_second_upper_estimate": 0.0017304247491323527
+      "required_token_positions_per_second": 28869.101582969968,
+      "required_pages_per_second_upper_estimate": 0.00173065855933328
     },
     {
       "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": 28865.201396878783,
-      "required_pages_per_second_upper_estimate": 0.0022441424440044133
+      "required_token_positions_per_second": 28869.101582969968,
+      "required_pages_per_second_upper_estimate": 0.0022444456663178974
     },
     {
       "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": 28865.201396878783,
-      "required_pages_per_second_upper_estimate": 0.37025181048060934
+      "required_token_positions_per_second": 28869.101582969968,
+      "required_pages_per_second_upper_estimate": 0.37030183788009347
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09767091942970296,
-    "producer_p95_seconds": 0.16575781500023368,
-    "gpu_update_seconds": 0.9578315293857468,
-    "producer_batches_per_second": 10.238462029834105,
-    "gpu_batches_per_second": 1.0440249347829422,
-    "producer_headroom": 9.806721744593899,
+    "producer_batch_seconds": 0.09439655206251629,
+    "producer_p95_seconds": 0.15547237500004485,
+    "gpu_update_seconds": 0.9577021273259746,
+    "producer_batches_per_second": 10.593607267961726,
+    "gpu_batches_per_second": 1.0441660005414486,
+    "producer_headroom": 10.14552021658285,
     "sustainable": true,
-    "buffer_seconds": 60.343386351302044,
+    "buffer_seconds": 60.3352340215364,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435296.9971199569,
+    "generation_bytes_per_second": 435419.66542711295,
     "upload_bytes_per_second": 5329050.240006242,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.40460238646536,
+    "estimated_full_shards_per_hour": 93.43092414960901,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 607400, "loss": 0.5584568977355957, "lr": 0.0001, "grad_norm": 0.13680674459670614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88905380803, "tokens_per_second": 152331.02339301273, "data_wait_s": 1.1147999884997262e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 607500, "loss": 0.5816423296928406, "lr": 0.0001, "grad_norm": 0.16900962176126952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88920076367, "tokens_per_second": 152590.53688233954, "data_wait_s": 1.2109000181226293e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 607600, "loss": 0.5491907000541687, "lr": 0.0001, "grad_norm": 0.1370757922205129, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88934813270, "tokens_per_second": 153153.35838071772, "data_wait_s": 1.163200022347155e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 607700, "loss": 0.5888853073120117, "lr": 0.0001, "grad_norm": 0.13579830744958846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88949481422, "tokens_per_second": 152801.69091184073, "data_wait_s": 7.930999800009886e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04232746c2e9f3bbcb14bba5f39728d939eb330f0dcb869fb6582b464f91478f
-size 68724
+oid sha256:cdc47628d22a29557ce35589b2af0a89377ffa8afc721b73a9f96dd9dadf7583
+size 70842