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": 76728.60237730146,
-      "required_pages_per_second_upper_estimate": 0.005308262747347651
+      "required_token_positions_per_second": 76931.50179054492,
+      "required_pages_per_second_upper_estimate": 0.0053222997995473345
     },
     {
       "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": 9591.075297162683,
-      "required_pages_per_second_upper_estimate": 0.004015582942494709
+      "required_token_positions_per_second": 9616.437723818115,
+      "required_pages_per_second_upper_estimate": 0.004026201660907644
     },
     {
       "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": 19182.150594325365,
-      "required_pages_per_second_upper_estimate": 0.003995136101468697
+      "required_token_positions_per_second": 19232.87544763623,
+      "required_pages_per_second_upper_estimate": 0.0040057007507090115
     },
     {
       "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": 28773.225891488048,
-      "required_pages_per_second_upper_estimate": 0.0017249109580226433
+      "required_token_positions_per_second": 28849.313171454345,
+      "required_pages_per_second_upper_estimate": 0.0017294722742780733
     },
     {
       "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": 28773.225891488048,
-      "required_pages_per_second_upper_estimate": 0.002236991753017777
+      "required_token_positions_per_second": 28849.313171454345,
+      "required_pages_per_second_upper_estimate": 0.0022429072043625813
     },
     {
       "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": 28773.225891488048,
-      "required_pages_per_second_upper_estimate": 0.3690720474530605
+      "required_token_positions_per_second": 28849.313171454345,
+      "required_pages_per_second_upper_estimate": 0.370048013384312
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05516035924219409,
-    "producer_p95_seconds": 0.13555075699991903,
-    "gpu_update_seconds": 0.9608933007466179,
-    "producer_batches_per_second": 18.12896097375423,
-    "gpu_batches_per_second": 1.0406982744317148,
-    "producer_headroom": 17.41999714917732,
+    "producer_batch_seconds": 0.05498347081250543,
+    "producer_p95_seconds": 0.13937091099978716,
+    "gpu_update_seconds": 0.9583590373776034,
+    "producer_batches_per_second": 18.18728401868294,
+    "gpu_batches_per_second": 1.043450273851792,
+    "producer_headroom": 17.429948004658055,
     "sustainable": true,
-    "buffer_seconds": 60.53627794703693,
+    "buffer_seconds": 60.376619354789014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 71,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 721300, "loss": 0.518498420715332, "lr": 5e-05, "grad_norm": 0.21344423893107728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105578467144, "tokens_per_second": 153266.1153677966, "data_wait_s": 1.3526999964597053e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721400, "loss": 0.5436562299728394, "lr": 5e-05, "grad_norm": 0.20138459009967297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105592944425, "tokens_per_second": 150507.00545782028, "data_wait_s": 1.8675999854167458e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721500, "loss": 0.564842939376831, "lr": 5e-05, "grad_norm": 0.22133629268959595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105607461004, "tokens_per_second": 151109.29452885545, "data_wait_s": 1.3805999969918048e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721600, "loss": 0.5053544044494629, "lr": 5e-05, "grad_norm": 0.20934199478775994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105622247717, "tokens_per_second": 153793.11842504656, "data_wait_s": 1.6168999991350574e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cdd97ae89aca984659440d276163c79b55d44ed090e9b5c99034cbcd6915b803
-size 31935
+oid sha256:0a1c47b0cc71f30615ebaffa07107a0993f02e6958b0e0a3b3b88e56bd4bd991
+size 34053