Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76388.50942967138,
-      "required_pages_per_second_upper_estimate": 0.0066211623924567874
+      "required_token_positions_per_second": 76480.22921346067,
+      "required_pages_per_second_upper_estimate": 0.006629112430853977
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9548.563678708922,
-      "required_pages_per_second_upper_estimate": 0.0038090648151862623
+      "required_token_positions_per_second": 9560.028651682584,
+      "required_pages_per_second_upper_estimate": 0.0038136383643220776
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19097.127357417845,
-      "required_pages_per_second_upper_estimate": 0.004417784316176299
+      "required_token_positions_per_second": 19120.057303365167,
+      "required_pages_per_second_upper_estimate": 0.004423088755617946
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28645.691036126766,
-      "required_pages_per_second_upper_estimate": 0.0017443722917744565
+      "required_token_positions_per_second": 28680.08595504775,
+      "required_pages_per_second_upper_estimate": 0.0017464667618805512
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28645.691036126766,
-      "required_pages_per_second_upper_estimate": 0.0022052245923090187
+      "required_token_positions_per_second": 28680.08595504775,
+      "required_pages_per_second_upper_estimate": 0.0022078724083787853
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28645.691036126766,
-      "required_pages_per_second_upper_estimate": 0.3556836117079947
+      "required_token_positions_per_second": 28680.08595504775,
+      "required_pages_per_second_upper_estimate": 0.35611068148860464
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.06634447944529143,
-    "producer_p95_seconds": 0.17216890799954854,
-    "gpu_update_seconds": 0.9651713399104767,
-    "producer_batches_per_second": 15.072844166704387,
-    "gpu_batches_per_second": 1.036085468609909,
-    "producer_headroom": 14.547877200639885,
+    "producer_batch_seconds": 0.06406504957034542,
+    "producer_p95_seconds": 0.17835820999971475,
+    "gpu_update_seconds": 0.9640138472156112,
+    "producer_batches_per_second": 15.609134882537925,
+    "gpu_batches_per_second": 1.0373294977954193,
+    "producer_headroom": 15.047422169822783,
     "sustainable": true,
-    "buffer_seconds": 60.80579441436003,
+    "buffer_seconds": 60.73287237458351,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 279000, "loss": 0.6319178156554699, "lr": 0.0003, "grad_norm": 0.17287498253791164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40838042164, "tokens_per_second": 152059.73354522864, "data_wait_s": 1.053100004355656e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 279100, "loss": 0.5977139621973038, "lr": 0.0003, "grad_norm": 0.15597813372850852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40852708011, "tokens_per_second": 151328.98478430815, "data_wait_s": 1.2958999832335394e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 279200, "loss": 0.6283826157450676, "lr": 0.0003, "grad_norm": 0.1496751405195431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40867476963, "tokens_per_second": 152577.3529347542, "data_wait_s": 1.2758000593748875e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 279300, "loss": 0.5953002441674471, "lr": 0.0003, "grad_norm": 0.1420977077373414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40882067628, "tokens_per_second": 150461.6564545454, "data_wait_s": 1.4311000086308923e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21422c67723d0ad77663b2e05bdacfbd4a98c743a7c0eeb396a906c132602dd6
-size 89121
+oid sha256:2a611aed18b237dcb0e34d8c883098e68d0012bc9846c55feb84905f2b8becb8
+size 91239