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": 76775.19613978494,
-      "required_pages_per_second_upper_estimate": 0.0053114862119486815
+      "required_token_positions_per_second": 76816.77737694119,
+      "required_pages_per_second_upper_estimate": 0.005314362898416914
     },
     {
       "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": 9596.899517473117,
-      "required_pages_per_second_upper_estimate": 0.0040180214219151375
+      "required_token_positions_per_second": 9602.097172117648,
+      "required_pages_per_second_upper_estimate": 0.004020197571375431
     },
     {
       "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": 19193.799034946234,
-      "required_pages_per_second_upper_estimate": 0.003997562164459987
+      "required_token_positions_per_second": 19204.194344235297,
+      "required_pages_per_second_upper_estimate": 0.003999727233242157
     },
     {
       "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": 28790.69855241935,
-      "required_pages_per_second_upper_estimate": 0.0017259584173662664
+      "required_token_positions_per_second": 28806.291516352947,
+      "required_pages_per_second_upper_estimate": 0.001726893191745008
     },
     {
       "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": 28790.69855241935,
-      "required_pages_per_second_upper_estimate": 0.0022383501755510724
+      "required_token_positions_per_second": 28806.291516352947,
+      "required_pages_per_second_upper_estimate": 0.00223956245991071
     },
     {
       "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": 28790.69855241935,
-      "required_pages_per_second_upper_estimate": 0.3692961679868056
+      "required_token_positions_per_second": 28806.291516352947,
+      "required_pages_per_second_upper_estimate": 0.3694961777857255
     }
   ],
   "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": 0.05591561326560068,
-    "producer_p95_seconds": 0.1314956420001181,
-    "gpu_update_seconds": 0.9603101484203714,
-    "producer_batches_per_second": 17.884092502928883,
-    "gpu_batches_per_second": 1.041330242781371,
-    "producer_headroom": 17.17427552585129,
+    "producer_batch_seconds": 0.05202940433593639,
+    "producer_p95_seconds": 0.12042292999967685,
+    "gpu_update_seconds": 0.9597903285920925,
+    "producer_batches_per_second": 19.2199009918187,
+    "gpu_batches_per_second": 1.0418942244051268,
+    "producer_headroom": 18.447075088445153,
     "sustainable": true,
-    "buffer_seconds": 60.4995393504834,
+    "buffer_seconds": 60.466790701301825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 58,
+    "upload_pages_per_commit": 59,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -6,3 +6,4 @@
 {"step": 800600, "loss": 0.5685425996780396, "lr": 3e-05, "grad_norm": 0.273817969088054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117184702810, "tokens_per_second": 153113.38106926592, "data_wait_s": 1.3907999800721882e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800700, "loss": 0.5554754137992859, "lr": 3e-05, "grad_norm": 0.27910189478111525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117199349780, "tokens_per_second": 152331.6607187394, "data_wait_s": 1.61310003932158e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800800, "loss": 0.5392715334892273, "lr": 3e-05, "grad_norm": 0.26546578503392076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117213898762, "tokens_per_second": 151639.58409232288, "data_wait_s": 1.4631999874836765e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 800900, "loss": 0.5483314394950867, "lr": 3e-05, "grad_norm": 0.27487786525222574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117228365568, "tokens_per_second": 150913.1377443111, "data_wait_s": 1.7877000118460273e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:233704eed7680829683e50fd1c0b744b5f72bdbede66de24002a7f19c87a2299
-size 17109
+oid sha256:1d599aaa9c4c19332aa01f865375bccd858f93b90b12ff54fd955fe0cb9c087d
+size 19227