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": 75772.45104969839,
-      "required_pages_per_second_upper_estimate": 0.00524211397992733
+      "required_token_positions_per_second": 77291.3914774091,
+      "required_pages_per_second_upper_estimate": 0.0053471978031437205
     },
     {
       "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": 9471.556381212298,
-      "required_pages_per_second_upper_estimate": 0.003965542868225059
+      "required_token_positions_per_second": 9661.423934676137,
+      "required_pages_per_second_upper_estimate": 0.0040450364479749906
     },
     {
       "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": 18943.112762424596,
-      "required_pages_per_second_upper_estimate": 0.0039453508249353095
+      "required_token_positions_per_second": 19322.847869352274,
+      "required_pages_per_second_upper_estimate": 0.004024439633420143
     },
     {
       "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": 28414.669143636893,
-      "required_pages_per_second_upper_estimate": 0.0017034160284734202
+      "required_token_positions_per_second": 28984.27180402841,
+      "required_pages_per_second_upper_estimate": 0.0017375628382309883
     },
     {
       "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": 28414.669143636893,
-      "required_pages_per_second_upper_estimate": 0.002209115542996808
+      "required_token_positions_per_second": 28984.27180402841,
+      "required_pages_per_second_upper_estimate": 0.002253399644424926
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28414.669143636893,
-      "required_pages_per_second_upper_estimate": 0.36447286647986676
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28984.27180402841,
+      "required_pages_per_second_upper_estimate": 0.3717791178156823
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11522439503869464,
-    "producer_p95_seconds": 0.31502199100214057,
-    "gpu_update_seconds": 0.9730185440568968,
-    "producer_batches_per_second": 8.678717728690874,
-    "gpu_batches_per_second": 1.027729642058626,
-    "producer_headroom": 8.444553288651573,
+    "producer_batch_seconds": 0.1278063406172123,
+    "producer_p95_seconds": 0.3259690260019852,
+    "gpu_update_seconds": 0.9538966577092792,
+    "producer_batches_per_second": 7.82433794106554,
+    "gpu_batches_per_second": 1.0483315901341295,
+    "producer_headroom": 7.463609810770322,
     "sustainable": true,
-    "buffer_seconds": 60.327149731527605,
+    "buffer_seconds": 60.09548943568459,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309621.9402706658,
-    "upload_bytes_per_second": 4123150.0361489994,
+    "generation_bytes_per_second": 309431.5116272109,
+    "upload_bytes_per_second": 4110176.0308090434,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.4376607522009,
+    "estimated_full_shards_per_hour": 66.39679919826742,
     "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": [
-    {
-      "seconds": 1.0960563930020726,
-      "bytes": 6111629,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.3281459399986488,
       "bytes": 11209095,
@@ -562,6 +556,12 @@
       "bytes": 6889142,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.881114518000686,
+      "bytes": 7983801,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -342,3 +342,4 @@
 {"step": 876700, "loss": 0.5308780670166016, "lr": 3e-05, "grad_norm": 0.30201104065519013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128324319271, "tokens_per_second": 136014.074377604, "data_wait_s": 8.424998668488115e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 876800, "loss": 0.5137549042701721, "lr": 3e-05, "grad_norm": 0.31649700093618344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128339018576, "tokens_per_second": 153615.26803584737, "data_wait_s": 1.62739961524494e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 876900, "loss": 0.572022557258606, "lr": 3e-05, "grad_norm": 0.3000676411673048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128353810830, "tokens_per_second": 153432.34944826318, "data_wait_s": 1.6711004718672484e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 877000, "loss": 0.5473411083221436, "lr": 3e-05, "grad_norm": 0.30369529390738514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128368665910, "tokens_per_second": 154402.80171194155, "data_wait_s": 1.091499871108681e-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:b6b6d21e00928bb23e43ee8d6cdb33162ba975880a653ac2a7a35d7513a81619
-size 734238
+oid sha256:3549caac6b794a19e1a1cf246509bd6927d2708ab90492986c0e68b4d15ecfba
+size 736356