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": 75039.92905469185,
-      "required_pages_per_second_upper_estimate": 0.006504270863499274
+      "required_token_positions_per_second": 75275.69562259261,
+      "required_pages_per_second_upper_estimate": 0.006524706512060009
     },
     {
       "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": 9379.991131836481,
-      "required_pages_per_second_upper_estimate": 0.0037418187058546677
+      "required_token_positions_per_second": 9409.461952824076,
+      "required_pages_per_second_upper_estimate": 0.0037535750569746597
     },
     {
       "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": 18759.982263672962,
-      "required_pages_per_second_upper_estimate": 0.004339791732289414
+      "required_token_positions_per_second": 18818.923905648153,
+      "required_pages_per_second_upper_estimate": 0.004353426843822905
     },
     {
       "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": 28139.97339550944,
-      "required_pages_per_second_upper_estimate": 0.0017135767407562641
+      "required_token_positions_per_second": 28228.38585847223,
+      "required_pages_per_second_upper_estimate": 0.0017189605958863542
     },
     {
       "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": 28139.97339550944,
-      "required_pages_per_second_upper_estimate": 0.002166293048418269
+      "required_token_positions_per_second": 28228.38585847223,
+      "required_pages_per_second_upper_estimate": 0.002173099284558506
     },
     {
       "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": 28139.97339550944,
-      "required_pages_per_second_upper_estimate": 0.3494042911395935
+      "required_token_positions_per_second": 28228.38585847223,
+      "required_pages_per_second_upper_estimate": 0.3505020780321123
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.05663709317181542,
-    "producer_p95_seconds": 0.1854400149986759,
-    "gpu_update_seconds": 0.9825169203753422,
-    "producer_batches_per_second": 17.656273371346582,
-    "gpu_batches_per_second": 1.0177941766315626,
-    "producer_headroom": 17.347587338120604,
+    "producer_batch_seconds": 0.10489945596876282,
+    "producer_p95_seconds": 0.20264205700004823,
+    "gpu_update_seconds": 0.9794396370595864,
+    "producer_batches_per_second": 9.532937904823664,
+    "gpu_batches_per_second": 1.0209919653671957,
+    "producer_headroom": 9.336937241612063,
     "sustainable": true,
-    "buffer_seconds": 60.91604906327122,
+    "buffer_seconds": 60.72525749769436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394326.17851746845,
+    "generation_bytes_per_second": 391621.7650537211,
     "upload_bytes_per_second": 5928699.59931897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.61321846621551,
+    "estimated_full_shards_per_hour": 84.03291429241871,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -91,3 +91,4 @@
 {"step": 49100, "loss": 0.6665434390306473, "lr": 0.0003, "grad_norm": 0.1404694792377278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7185626821, "tokens_per_second": 151023.4391343018, "data_wait_s": 7.995000487426296e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49200, "loss": 0.7101405002176762, "lr": 0.0003, "grad_norm": 0.1457871580642793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7200327105, "tokens_per_second": 143859.85854952806, "data_wait_s": 1.0122999810846522e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 49300, "loss": 0.6631663143634796, "lr": 0.0003, "grad_norm": 0.1397440728156991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7215087862, "tokens_per_second": 151228.84642449685, "data_wait_s": 7.808999725966714e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 49400, "loss": 0.6320028305053711, "lr": 0.0003, "grad_norm": 0.1333903930589001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7229811511, "tokens_per_second": 150615.8563482586, "data_wait_s": 1.1651000022538938e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:996ecfdffecb2fdcd37d8248077568efbffada849a94ef58004c7cf738610cba
-size 197922
+oid sha256:8bf58cb9aaa6563f8c5263f9f791a3f1b8fa7ece763917274555936a3d55af38
+size 200040