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": 77407.83355310651,
-      "required_pages_per_second_upper_estimate": 0.005355253536123307
+      "required_token_positions_per_second": 77349.77899716243,
+      "required_pages_per_second_upper_estimate": 0.005351237187237967
     },
     {
       "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": 9675.979194138314,
-      "required_pages_per_second_upper_estimate": 0.004051130431163423
+      "required_token_positions_per_second": 9668.722374645304,
+      "required_pages_per_second_upper_estimate": 0.0040480921523813225
     },
     {
       "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": 19351.958388276627,
-      "required_pages_per_second_upper_estimate": 0.004030502586816077
+      "required_token_positions_per_second": 19337.444749290607,
+      "required_pages_per_second_upper_estimate": 0.004027479778565688
     },
     {
       "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": 29027.937582414943,
-      "required_pages_per_second_upper_estimate": 0.001740180534971478
+      "required_token_positions_per_second": 29006.16712393591,
+      "required_pages_per_second_upper_estimate": 0.0017388754292272251
     },
     {
       "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": 29027.937582414943,
-      "required_pages_per_second_upper_estimate": 0.002256794466629018
+      "required_token_positions_per_second": 29006.16712393591,
+      "required_pages_per_second_upper_estimate": 0.0022551019092404
     },
     {
       "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": 29027.937582414943,
-      "required_pages_per_second_upper_estimate": 0.3723392155361648
+      "required_token_positions_per_second": 29006.16712393591,
+      "required_pages_per_second_upper_estimate": 0.37205996747009284
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,14 +158,14 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04676419059375547,
-    "producer_p95_seconds": 0.134203406999859,
-    "gpu_update_seconds": 0.9524617421235291,
-    "producer_batches_per_second": 21.383883422405095,
-    "gpu_batches_per_second": 1.0499109368639665,
-    "producer_headroom": 20.36733085787041,
+    "producer_batch_seconds": 0.055237927257813624,
+    "producer_p95_seconds": 0.14168005800001993,
+    "gpu_update_seconds": 0.953176608335296,
+    "producer_batches_per_second": 18.103503329020118,
+    "gpu_batches_per_second": 1.0491235215543948,
+    "producer_headroom": 17.255835902142135,
     "sustainable": true,
-    "buffer_seconds": 60.00508975378233,
+    "buffer_seconds": 60.05012632512365,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 647300, "loss": 0.5541459321975708, "lr": 0.0001, "grad_norm": 0.19038575680539732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94747831707, "tokens_per_second": 152961.1084661129, "data_wait_s": 1.2200999663036782e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 647400, "loss": 0.4948424994945526, "lr": 0.0001, "grad_norm": 0.17461354103358048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94762596298, "tokens_per_second": 154340.7319993418, "data_wait_s": 1.4572000054613454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 647500, "loss": 0.5511292815208435, "lr": 0.0001, "grad_norm": 0.17853451125416817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94777372465, "tokens_per_second": 154982.65392728505, "data_wait_s": 1.3123999906383688e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 647600, "loss": 0.57598477602005, "lr": 0.0001, "grad_norm": 0.17728878264916242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94792018838, "tokens_per_second": 153620.90615440658, "data_wait_s": 1.6302999938488938e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3581881b946f97d83b1054e50ef9c161ec7cee2cffa7ba550bc4ef276d9895ea
-size 53115
+oid sha256:4631ca276d2a16519b9432d82120907128b897983dbd6c2a357c62b06655adbd
+size 55233