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": 76769.29696108651,
-      "required_pages_per_second_upper_estimate": 0.005311078093078344
+      "required_token_positions_per_second": 76859.32081709134,
+      "required_pages_per_second_upper_estimate": 0.005317306152320883
     },
     {
       "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": 9596.162120135814,
-      "required_pages_per_second_upper_estimate": 0.0040177126890486164
+      "required_token_positions_per_second": 9607.415102136418,
+      "required_pages_per_second_upper_estimate": 0.004022424077623283
     },
     {
       "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": 19192.324240271628,
-      "required_pages_per_second_upper_estimate": 0.003997255003622217
+      "required_token_positions_per_second": 19214.830204272836,
+      "required_pages_per_second_upper_estimate": 0.004001942402401486
     },
     {
       "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": 28788.48636040744,
-      "required_pages_per_second_upper_estimate": 0.001725825799832974
+      "required_token_positions_per_second": 28822.245306409255,
+      "required_pages_per_second_upper_estimate": 0.001727849597098857
     },
     {
       "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": 28788.48636040744,
-      "required_pages_per_second_upper_estimate": 0.002238178187352551
+      "required_token_positions_per_second": 28822.245306409255,
+      "required_pages_per_second_upper_estimate": 0.002240802796914281
     },
     {
       "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": 28788.48636040744,
-      "required_pages_per_second_upper_estimate": 0.36926779236294355
+      "required_token_positions_per_second": 28822.245306409255,
+      "required_pages_per_second_upper_estimate": 0.3697008158747227
     }
   ],
   "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": 0.05223049150782444,
-    "producer_p95_seconds": 0.13301803500007736,
-    "gpu_update_seconds": 0.9603839414782174,
-    "producer_batches_per_second": 19.1459044541098,
-    "gpu_batches_per_second": 1.041250230049459,
-    "producer_headroom": 18.387419182803328,
+    "producer_batch_seconds": 0.04724863498438481,
+    "producer_p95_seconds": 0.12650885600010042,
+    "gpu_update_seconds": 0.9592590620915944,
+    "producer_batches_per_second": 21.164632593735032,
+    "gpu_batches_per_second": 1.042471256742233,
+    "producer_headroom": 20.302365611379457,
     "sustainable": true,
-    "buffer_seconds": 60.50418831312769,
+    "buffer_seconds": 60.433320911770444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 71,
+    "upload_pages_per_commit": 74,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -15,3 +15,4 @@
 {"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}
 {"step": 721700, "loss": 0.5054534673690796, "lr": 5e-05, "grad_norm": 0.21090170174195838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105636903287, "tokens_per_second": 152603.45544524852, "data_wait_s": 1.4651000128651503e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721800, "loss": 0.597263753414154, "lr": 5e-05, "grad_norm": 0.21663511348799927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105651647851, "tokens_per_second": 153471.2782438447, "data_wait_s": 1.55860002450936e-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:e62cc85734bf863236bf36423c58b9e1da4d2cd19e022ca5e20d98031987143e
-size 36171
+oid sha256:7ef78b4d5f604e066ddb296d4aceb991ed6a48a7ea9501126e524c242e09f8de
+size 38289