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": 76591.91576296306,
-      "required_pages_per_second_upper_estimate": 0.00529880645021108
+      "required_token_positions_per_second": 76908.38335048531,
+      "required_pages_per_second_upper_estimate": 0.005320700412221839
     },
     {
       "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": 9573.989470370383,
-      "required_pages_per_second_upper_estimate": 0.004008429463609409
+      "required_token_positions_per_second": 9613.547918810664,
+      "required_pages_per_second_upper_estimate": 0.004024991759897015
     },
     {
       "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": 19147.978940740766,
-      "required_pages_per_second_upper_estimate": 0.00398801904719413
+      "required_token_positions_per_second": 19227.095837621328,
+      "required_pages_per_second_upper_estimate": 0.004004497010361473
     },
     {
       "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": 28721.96841111115,
-      "required_pages_per_second_upper_estimate": 0.0017218381503396889
+      "required_token_positions_per_second": 28840.64375643199,
+      "required_pages_per_second_upper_estimate": 0.0017289525560849133
     },
     {
       "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": 28721.96841111115,
-      "required_pages_per_second_upper_estimate": 0.002233006709376302
+      "required_token_positions_per_second": 28840.64375643199,
+      "required_pages_per_second_upper_estimate": 0.00224223319547732
     },
     {
       "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": 28721.96841111115,
-      "required_pages_per_second_upper_estimate": 0.3684145715307801
+      "required_token_positions_per_second": 28840.64375643199,
+      "required_pages_per_second_upper_estimate": 0.3699368114368978
     }
   ],
   "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": 0.04355810394532078,
-    "producer_p95_seconds": 0.11743981199992959,
-    "gpu_update_seconds": 0.9626081194805687,
-    "producer_batches_per_second": 22.957840434361348,
-    "gpu_batches_per_second": 1.0388443435731751,
-    "producer_headroom": 22.09940360785554,
+    "producer_batch_seconds": 0.0530859903047034,
+    "producer_p95_seconds": 0.12853050700005042,
+    "gpu_update_seconds": 0.9586471173631134,
+    "producer_batches_per_second": 18.837361689217662,
+    "gpu_batches_per_second": 1.0431367099403932,
+    "producer_headroom": 18.058382482094856,
     "sustainable": true,
-    "buffer_seconds": 60.644311527275825,
+    "buffer_seconds": 60.39476839387614,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 66,
+    "upload_pages_per_commit": 65,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -6,3 +6,4 @@
 {"step": 720600, "loss": 0.5312745571136475, "lr": 5e-05, "grad_norm": 0.21549012287903854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105475828610, "tokens_per_second": 152396.13794055022, "data_wait_s": 1.341500001217355e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720700, "loss": 0.519378662109375, "lr": 5e-05, "grad_norm": 0.22352734966768664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105490324667, "tokens_per_second": 150780.89103931733, "data_wait_s": 1.3140999953975552e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720800, "loss": 0.495766818523407, "lr": 5e-05, "grad_norm": 0.21895161647899114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105504891856, "tokens_per_second": 151486.92475880412, "data_wait_s": 1.476499983255053e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720900, "loss": 0.5261343717575073, "lr": 5e-05, "grad_norm": 0.21882450318529473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105519523252, "tokens_per_second": 151969.8927038483, "data_wait_s": 1.5592000181641197e-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 sha256:071a8b69eafbe2d3587f9d2a920fc74fd1b3850064c160a99c2c7aafe9bfee5e
-size 17109
+oid sha256:a7ae3719d8c37ca56cf9faf60c9b2e03122c3d9bc289e75b25fc0320d52ba571
+size 19227