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": 75818.29060722202,
-      "required_pages_per_second_upper_estimate": 0.005245285266879242
+      "required_token_positions_per_second": 75501.3095122896,
+      "required_pages_per_second_upper_estimate": 0.005223355779234346
     },
     {
       "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": 9477.286325902753,
-      "required_pages_per_second_upper_estimate": 0.003967941876412101
+      "required_token_positions_per_second": 9437.6636890362,
+      "required_pages_per_second_upper_estimate": 0.003951352705770822
     },
     {
       "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": 18954.572651805505,
-      "required_pages_per_second_upper_estimate": 0.003947737617675746
+      "required_token_positions_per_second": 18875.3273780724,
+      "required_pages_per_second_upper_estimate": 0.003931232916995545
     },
     {
       "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": 28431.85897770826,
-      "required_pages_per_second_upper_estimate": 0.0017044465327786416
+      "required_token_positions_per_second": 28312.991067108604,
+      "required_pages_per_second_upper_estimate": 0.0016973205830389825
     },
     {
       "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": 28431.85897770826,
-      "required_pages_per_second_upper_estimate": 0.0022104519769857675
+      "required_token_positions_per_second": 28312.991067108604,
+      "required_pages_per_second_upper_estimate": 0.002201210519781327
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28431.85897770826,
-      "required_pages_per_second_upper_estimate": 0.36469335921432844
+      "required_token_positions_per_second": 28312.991067108604,
+      "required_pages_per_second_upper_estimate": 0.3631686492875746
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12019547482790927,
-    "producer_p95_seconds": 0.25744350599779864,
-    "gpu_update_seconds": 0.9724302593677453,
-    "producer_batches_per_second": 8.319780769049395,
-    "gpu_batches_per_second": 1.028351380848823,
-    "producer_headroom": 8.090406571129481,
+    "producer_batch_seconds": 0.3967515599530316,
+    "producer_p95_seconds": 0.26873104799960856,
+    "gpu_update_seconds": 0.9765128641642836,
+    "producer_batches_per_second": 2.520468980937044,
+    "gpu_batches_per_second": 1.0240520495916017,
+    "producer_headroom": 2.461270383612066,
     "sustainable": true,
-    "buffer_seconds": 60.29067608080021,
+    "buffer_seconds": 60.54379757818558,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310137.55329473515,
+    "generation_bytes_per_second": 290529.505912798,
     "upload_bytes_per_second": 4783958.9458460165,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.54829930430928,
+    "estimated_full_shards_per_hour": 62.340868788127466,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 911800, "loss": 0.5464526414871216, "lr": 3e-05, "grad_norm": 0.3241595484439124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133460996453, "tokens_per_second": 152485.41767541075, "data_wait_s": 1.4429999282583594e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911900, "loss": 0.5535483956336975, "lr": 3e-05, "grad_norm": 0.3318998368641891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133475655567, "tokens_per_second": 152193.08449663408, "data_wait_s": 1.050200080499053e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912000, "loss": 0.554841160774231, "lr": 3e-05, "grad_norm": 0.32315140536455805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133490399358, "tokens_per_second": 153303.78350690423, "data_wait_s": 1.52729990077205e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 912100, "loss": 0.5340381860733032, "lr": 3e-05, "grad_norm": 0.33642669476817466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133505129415, "tokens_per_second": 153514.19093113282, "data_wait_s": 1.6370999219361693e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c38b9b0d2902f1de50b6016e6e75e4d507c10a8dc50cb72ed4e8b1b09a791ae
-size 575940
+oid sha256:e95af9d8b56beb2d9cf9680b203aa3c1dbd113b92707686fbbbc06bc2a6944e3
+size 578058