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": 77202.54544142184,
-      "required_pages_per_second_upper_estimate": 0.0053410512282229166
+      "required_token_positions_per_second": 76326.09737649353,
+      "required_pages_per_second_upper_estimate": 0.005280416517451485
     },
     {
       "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": 9650.31818017773,
-      "required_pages_per_second_upper_estimate": 0.004040386700481033
+      "required_token_positions_per_second": 9540.762172061692,
+      "required_pages_per_second_upper_estimate": 0.003994517887672451
     },
     {
       "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": 19300.63636035546,
-      "required_pages_per_second_upper_estimate": 0.00401981356185299
+      "required_token_positions_per_second": 19081.524344123383,
+      "required_pages_per_second_upper_estimate": 0.003974178307244295
     },
     {
       "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": 28950.95454053319,
-      "required_pages_per_second_upper_estimate": 0.0017355655191569656
+      "required_token_positions_per_second": 28622.28651618507,
+      "required_pages_per_second_upper_estimate": 0.0017158623729443111
     },
     {
       "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": 28950.95454053319,
-      "required_pages_per_second_upper_estimate": 0.002250809373735327
+      "required_token_positions_per_second": 28622.28651618507,
+      "required_pages_per_second_upper_estimate": 0.0022252568805001182
     },
     {
       "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": 28950.95454053319,
-      "required_pages_per_second_upper_estimate": 0.371351759732856
+      "required_token_positions_per_second": 28622.28651618507,
+      "required_pages_per_second_upper_estimate": 0.36713595921274833
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08093316656248817,
-    "producer_p95_seconds": 0.21716480599934584,
-    "gpu_update_seconds": 0.9549944186223992,
-    "producer_batches_per_second": 12.355873895381372,
-    "gpu_batches_per_second": 1.047126538647757,
-    "producer_headroom": 11.799790607291412,
+    "producer_batch_seconds": 0.12540559882054936,
+    "producer_p95_seconds": 0.2516024539982027,
+    "gpu_update_seconds": 0.9659605630865954,
+    "producer_batches_per_second": 7.974125632388726,
+    "gpu_batches_per_second": 1.035238950961555,
+    "producer_headroom": 7.702690885985467,
     "sustainable": true,
-    "buffer_seconds": 60.16464837321115,
+    "buffer_seconds": 60.85551547445551,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307340.30772477057,
-    "upload_bytes_per_second": 4613119.060990557,
+    "generation_bytes_per_second": 320171.5846974067,
+    "upload_bytes_per_second": 4492336.206827352,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.9480755215391,
+    "estimated_full_shards_per_hour": 68.70136886302615,
     "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": 1.1024076169996988,
-      "bytes": 12461960,
-      "pages": 47,
-      "shards": 3
-    },
     {
       "seconds": 1.0951313269997627,
       "bytes": 6163611,
@@ -562,6 +556,12 @@
       "bytes": 10144494,
       "pages": 44,
       "shards": 3
+    },
+    {
+      "seconds": 2.5764055170002393,
+      "bytes": 7638759,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 458900, "loss": 0.5333598256111145, "lr": 0.0003, "grad_norm": 0.1435300061299573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67168875836, "tokens_per_second": 153136.1474479134, "data_wait_s": 8.0740028352011e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 459000, "loss": 0.6018263697624207, "lr": 0.0003, "grad_norm": 0.1608775746947532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67183645713, "tokens_per_second": 153250.3294418098, "data_wait_s": 9.354000212624669e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 459100, "loss": 0.6053174734115601, "lr": 0.0003, "grad_norm": 0.16764479636843854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67198225526, "tokens_per_second": 151924.2175549408, "data_wait_s": 1.532199894427322e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 459200, "loss": 0.5952470898628235, "lr": 0.0003, "grad_norm": 0.1535267059201879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67212758806, "tokens_per_second": 150534.14906293093, "data_wait_s": 1.428700124961324e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3ef9ab5a4c8ff5412822c787593fb851a2b42e83db020b98ae22f75696eee25
-size 402816
+oid sha256:7aee20d2363ac7a0ddbca9fd895baeeeafdbbc911fae74a6a0947e487fb15c09
+size 404934