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": 77531.43049493898,
-      "required_pages_per_second_upper_estimate": 0.005363804259343693
+      "required_token_positions_per_second": 77528.61110197607,
+      "required_pages_per_second_upper_estimate": 0.005363609207196626
     },
     {
       "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": 9691.428811867372,
-      "required_pages_per_second_upper_estimate": 0.0040575988634818745
+      "required_token_positions_per_second": 9691.076387747009,
+      "required_pages_per_second_upper_estimate": 0.004057451310862131
     },
     {
       "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": 19382.857623734744,
-      "required_pages_per_second_upper_estimate": 0.004036938082694366
+      "required_token_positions_per_second": 19382.152775494018,
+      "required_pages_per_second_upper_estimate": 0.004036791281393921
     },
     {
       "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": 29074.286435602116,
-      "required_pages_per_second_upper_estimate": 0.001742959077949448
+      "required_token_positions_per_second": 29073.229163241027,
+      "required_pages_per_second_upper_estimate": 0.001742895696085762
     },
     {
       "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": 29074.286435602116,
-      "required_pages_per_second_upper_estimate": 0.0022603978860972606
+      "required_token_positions_per_second": 29073.229163241027,
+      "required_pages_per_second_upper_estimate": 0.0022603156878216357
     },
     {
       "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": 29074.286435602116,
-      "required_pages_per_second_upper_estimate": 0.3729337288593286
+      "required_token_positions_per_second": 29073.229163241027,
+      "required_pages_per_second_upper_estimate": 0.37292016730469113
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14333037889815614,
-    "producer_p95_seconds": 0.34880387199518736,
-    "gpu_update_seconds": 0.9509433726340538,
-    "producer_batches_per_second": 6.976887996023182,
-    "gpu_batches_per_second": 1.0515873276765813,
-    "producer_headroom": 6.63462540142833,
+    "producer_batch_seconds": 0.09087288124982251,
+    "producer_p95_seconds": 0.3445419660056359,
+    "gpu_update_seconds": 0.9509779544873183,
+    "producer_batches_per_second": 11.004383114593422,
+    "gpu_batches_per_second": 1.051549087212132,
+    "producer_headroom": 10.464925744710836,
     "sustainable": true,
-    "buffer_seconds": 60.860375848579444,
+    "buffer_seconds": 60.86258908718837,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333269.8030870621,
+    "generation_bytes_per_second": 333314.93841899995,
     "upload_bytes_per_second": 4916723.908153679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.51194161852739,
+    "estimated_full_shards_per_hour": 71.5216266100645,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 962800, "loss": 0.514863908290863, "lr": 3e-05, "grad_norm": 0.34611537754554306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140928191675, "tokens_per_second": 153670.39868493073, "data_wait_s": 1.0747004125732929e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962900, "loss": 0.5359106063842773, "lr": 3e-05, "grad_norm": 0.3646046451258107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140942857852, "tokens_per_second": 152803.81913517698, "data_wait_s": 9.566996595822275e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 963000, "loss": 0.5157766342163086, "lr": 3e-05, "grad_norm": 0.35943911031823833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140957303381, "tokens_per_second": 150299.34250841028, "data_wait_s": 1.4718003512825817e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 963100, "loss": 0.5452260375022888, "lr": 3e-05, "grad_norm": 0.35641047305290147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140971744535, "tokens_per_second": 133189.05502228468, "data_wait_s": 1.3133001630194485e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5937cb3399bb7cd257937a293b1c293139c8003917f03223cc29d6db4685ae80
-size 810486
+oid sha256:bc053e7a56778b30fa0c51b09eaa947e8c3992f6d1a43dcf80acd9a9be139831
+size 812604