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": 76927.42119940829,
-      "required_pages_per_second_upper_estimate": 0.005322017494784227
+      "required_token_positions_per_second": 76673.10835950509,
+      "required_pages_per_second_upper_estimate": 0.005304423542432636
     },
     {
       "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": 9615.927649926036,
-      "required_pages_per_second_upper_estimate": 0.004025988103620585
+      "required_token_positions_per_second": 9584.138544938136,
+      "required_pages_per_second_upper_estimate": 0.00401267866919415
     },
     {
       "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": 19231.85529985207,
-      "required_pages_per_second_upper_estimate": 0.004005488280828678
+      "required_token_positions_per_second": 19168.277089876272,
+      "required_pages_per_second_upper_estimate": 0.0039922466163608665
     },
     {
       "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": 28847.78294977811,
-      "required_pages_per_second_upper_estimate": 0.001729380539825096
+      "required_token_positions_per_second": 28752.415634814406,
+      "required_pages_per_second_upper_estimate": 0.0017236634149104806
     },
     {
       "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": 28847.78294977811,
-      "required_pages_per_second_upper_estimate": 0.0022427882363579878
+      "required_token_positions_per_second": 28752.415634814406,
+      "required_pages_per_second_upper_estimate": 0.0022353738470960467
     },
     {
       "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": 28847.78294977811,
-      "required_pages_per_second_upper_estimate": 0.3700283853436732
+      "required_token_positions_per_second": 28752.415634814406,
+      "required_pages_per_second_upper_estimate": 0.3688051158247637
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05411437682816711,
-    "producer_p95_seconds": 0.18370997800047917,
-    "gpu_update_seconds": 0.9584098732347355,
-    "producer_batches_per_second": 18.479377544628573,
-    "gpu_batches_per_second": 1.0433949272923215,
-    "producer_headroom": 17.710817890004286,
+    "producer_batch_seconds": 0.0768606520313142,
+    "producer_p95_seconds": 0.19771702500111132,
+    "gpu_update_seconds": 0.961588770528305,
+    "producer_batches_per_second": 13.010558375078379,
+    "gpu_batches_per_second": 1.0399455886434608,
+    "producer_headroom": 12.51080683177836,
     "sustainable": true,
-    "buffer_seconds": 60.379822013788335,
+    "buffer_seconds": 60.580092543283214,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 254930.8164109739,
+    "generation_bytes_per_second": 253704.16431534512,
     "upload_bytes_per_second": 4814753.355319817,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.70221871611512,
+    "estimated_full_shards_per_hour": 54.43900773139253,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 853500, "loss": 0.5550978779792786, "lr": 3e-05, "grad_norm": 0.2965777197136554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124928502481, "tokens_per_second": 137854.85290137946, "data_wait_s": 1.142899964179378e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 853600, "loss": 0.5497733354568481, "lr": 3e-05, "grad_norm": 0.2856565265019204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124943209301, "tokens_per_second": 153404.2340199271, "data_wait_s": 1.5001000065240078e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 853700, "loss": 0.5567955374717712, "lr": 3e-05, "grad_norm": 0.30432902070128004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124957907790, "tokens_per_second": 152826.08968260873, "data_wait_s": 1.6863999917404726e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 853800, "loss": 0.5145642757415771, "lr": 3e-05, "grad_norm": 0.2945418334246771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124972555962, "tokens_per_second": 152459.7652867361, "data_wait_s": 1.4790999557590112e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bbba557418116ad5a52f280694d28e300bcc1cc9c6ab55fd709d6d6cc28af2ae
-size 238947
+oid sha256:8bda1da4e1779f91be50b8352349d51813aab271a30d93800ae6e1f1b28d150b
+size 241065