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": 77505.602358435,
-      "required_pages_per_second_upper_estimate": 0.005362017408930816
+      "required_token_positions_per_second": 77539.51724529828,
+      "required_pages_per_second_upper_estimate": 0.005364363719497396
     },
     {
       "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": 9688.200294804376,
-      "required_pages_per_second_upper_estimate": 0.0040562471508067
+      "required_token_positions_per_second": 9692.439655662285,
+      "required_pages_per_second_upper_estimate": 0.004058022082669986
     },
     {
       "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": 19376.40058960875,
-      "required_pages_per_second_upper_estimate": 0.004035593252769362
+      "required_token_positions_per_second": 19384.87931132457,
+      "required_pages_per_second_upper_estimate": 0.00403735914690384
     },
     {
       "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": 29064.600884413125,
-      "required_pages_per_second_upper_estimate": 0.0017423784439446481
+      "required_token_positions_per_second": 29077.31896698686,
+      "required_pages_per_second_upper_estimate": 0.0017431408735755559
     },
     {
       "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": 29064.600884413125,
-      "required_pages_per_second_upper_estimate": 0.002259644877094554
+      "required_token_positions_per_second": 29077.31896698686,
+      "required_pages_per_second_upper_estimate": 0.0022606336520737285
     },
     {
       "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": 29064.600884413125,
-      "required_pages_per_second_upper_estimate": 0.3728094930082108
+      "required_token_positions_per_second": 29077.31896698686,
+      "required_pages_per_second_upper_estimate": 0.37297262691585353
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,14 +158,14 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0456452496874995,
-    "producer_p95_seconds": 0.1339244260002488,
-    "gpu_update_seconds": 0.9512602670841138,
-    "producer_batches_per_second": 21.908084780919975,
-    "gpu_batches_per_second": 1.0512370111549887,
-    "producer_headroom": 20.840290579999344,
+    "producer_batch_seconds": 0.048908546937539654,
+    "producer_p95_seconds": 0.12753010100004758,
+    "gpu_update_seconds": 0.9508441968597708,
+    "producer_batches_per_second": 20.446324060231937,
+    "gpu_batches_per_second": 1.051697011248078,
+    "producer_headroom": 19.441268579785845,
     "sustainable": true,
-    "buffer_seconds": 60.880657093383284,
+    "buffer_seconds": 60.854028599025334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 68,

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 526300, "loss": 0.6389694213867188, "lr": 0.0003, "grad_norm": 0.16326968418129384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77032790042, "tokens_per_second": 153197.9701533869, "data_wait_s": 1.2168000012025004e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526400, "loss": 0.5397037863731384, "lr": 0.0003, "grad_norm": 0.15235547283058473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77047436084, "tokens_per_second": 153645.020126853, "data_wait_s": 1.8833000012818957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526500, "loss": 0.5814909338951111, "lr": 0.0003, "grad_norm": 0.17019092939085942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77062021391, "tokens_per_second": 153103.6932522806, "data_wait_s": 1.4146000012260629e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526600, "loss": 0.5899880528450012, "lr": 0.0003, "grad_norm": 0.15871447608584385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77076707540, "tokens_per_second": 154316.9364779082, "data_wait_s": 1.3916999705543276e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe5b21c85e8718dee26a3b6d7bc5daa349506f1459913e5ab6c35d598880a460
-size 31935
+oid sha256:b24f517136bbdd5331c42c13aba04230fdfaf31bda1815a23344f233bf2df8bf
+size 34053