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": 76364.64985359696,
-      "required_pages_per_second_upper_estimate": 0.005283083667271571
+      "required_token_positions_per_second": 77221.96522428136,
+      "required_pages_per_second_upper_estimate": 0.005342394733861242
     },
     {
       "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": 9545.58123169962,
-      "required_pages_per_second_upper_estimate": 0.0039965355273094425
+      "required_token_positions_per_second": 9652.74565303517,
+      "required_pages_per_second_upper_estimate": 0.004041403032675046
     },
     {
       "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": 19091.16246339924,
-      "required_pages_per_second_upper_estimate": 0.003976185673315158
+      "required_token_positions_per_second": 19305.49130607034,
+      "required_pages_per_second_upper_estimate": 0.004020824719011871
     },
     {
       "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": 28636.74369509886,
-      "required_pages_per_second_upper_estimate": 0.0017167290587453588
+      "required_token_positions_per_second": 28958.23695910551,
+      "required_pages_per_second_upper_estimate": 0.0017360020890307669
     },
     {
       "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": 28636.74369509886,
-      "required_pages_per_second_upper_estimate": 0.0022263808625702557
+      "required_token_positions_per_second": 28958.23695910551,
+      "required_pages_per_second_upper_estimate": 0.002251375549747351
     },
     {
       "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": 28636.74369509886,
-      "required_pages_per_second_upper_estimate": 0.3673214003809451
+      "required_token_positions_per_second": 28958.23695910551,
+      "required_pages_per_second_upper_estimate": 0.3714451707790499
     }
   ],
   "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": 0.05347612466410823,
-    "producer_p95_seconds": 0.16805529100020067,
-    "gpu_update_seconds": 0.9654729006333188,
-    "producer_batches_per_second": 18.699933966441172,
-    "gpu_batches_per_second": 1.0357618523979621,
-    "producer_headroom": 18.054279488231483,
+    "producer_batch_seconds": 0.06704962596878516,
+    "producer_p95_seconds": 0.16964902199924836,
+    "gpu_update_seconds": 0.9547542565883479,
+    "producer_batches_per_second": 14.914326300128032,
+    "gpu_batches_per_second": 1.0473899363102397,
+    "producer_headroom": 14.239516519194783,
     "sustainable": true,
-    "buffer_seconds": 60.82479273989909,
+    "buffer_seconds": 60.14951816506592,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419033.50047501427,
-    "upload_bytes_per_second": 4439422.224069715,
+    "generation_bytes_per_second": 436921.7244424384,
+    "upload_bytes_per_second": 4455721.912451897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.91483460128613,
+    "estimated_full_shards_per_hour": 93.75323104815354,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 7339507,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0799733499998183,
+      "bytes": 6017108,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 848600, "loss": 0.5495550632476807, "lr": 3e-05, "grad_norm": 0.2998802889979366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124211473077, "tokens_per_second": 151325.2296496637, "data_wait_s": 1.6713000150048174e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848700, "loss": 0.5220390558242798, "lr": 3e-05, "grad_norm": 0.30243580551810806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124226175902, "tokens_per_second": 152910.14577780204, "data_wait_s": 1.1047999578295276e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848800, "loss": 0.5162243247032166, "lr": 3e-05, "grad_norm": 0.2876559223096013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124240837254, "tokens_per_second": 152488.06106000085, "data_wait_s": 8.566999895265326e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 848900, "loss": 0.4928153157234192, "lr": 3e-05, "grad_norm": 0.289014001151626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124255399458, "tokens_per_second": 151503.57088511495, "data_wait_s": 1.4599000678572338e-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 sha256:1bd9a8964949c801f02744fb1e343c85127b3f22e29c9fa82d062f1083965e6b
-size 134382
+oid sha256:a92ae8860ff3067329961d337fcdd138a5bcdb726d9fb47800e291fc9c05133a
+size 136500