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": 76075.3752010543,
-      "required_pages_per_second_upper_estimate": 0.005263070975599001
+      "required_token_positions_per_second": 76248.49827842129,
+      "required_pages_per_second_upper_estimate": 0.005275048031792133
     },
     {
       "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": 9509.421900131787,
-      "required_pages_per_second_upper_estimate": 0.003981396370274698
+      "required_token_positions_per_second": 9531.062284802661,
+      "required_pages_per_second_upper_estimate": 0.003990456747433774
     },
     {
       "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": 19018.843800263574,
-      "required_pages_per_second_upper_estimate": 0.003961123602955397
+      "required_token_positions_per_second": 19062.124569605323,
+      "required_pages_per_second_upper_estimate": 0.003970137845818641
     },
     {
       "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": 28528.26570039536,
-      "required_pages_per_second_upper_estimate": 0.0017102259685991912
+      "required_token_positions_per_second": 28593.186854407984,
+      "required_pages_per_second_upper_estimate": 0.0017141178926533857
     },
     {
       "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": 28528.26570039536,
-      "required_pages_per_second_upper_estimate": 0.0022179471756263312
+      "required_token_positions_per_second": 28593.186854407984,
+      "required_pages_per_second_upper_estimate": 0.002222994509792835
     },
     {
       "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": 28528.26570039536,
-      "required_pages_per_second_upper_estimate": 0.3659299611394846
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28593.186854407984,
+      "required_pages_per_second_upper_estimate": 0.3667626999962543
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08939356210163396,
-    "producer_p95_seconds": 0.2542347839989816,
-    "gpu_update_seconds": 0.9691440864425501,
-    "producer_batches_per_second": 11.186487891186983,
-    "gpu_batches_per_second": 1.0318383138163831,
-    "producer_headroom": 10.841318587805059,
+    "producer_batch_seconds": 0.28394392831248183,
+    "producer_p95_seconds": 0.26178744599747006,
+    "gpu_update_seconds": 0.9669436338376437,
+    "producer_batches_per_second": 3.521822093337719,
+    "gpu_batches_per_second": 1.034186445833622,
+    "producer_headroom": 3.4054034526616714,
     "sustainable": true,
-    "buffer_seconds": 60.086933359438106,
+    "buffer_seconds": 60.91744893177155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367448.05649861845,
-    "upload_bytes_per_second": 4917165.025646714,
+    "generation_bytes_per_second": 399967.0588958948,
+    "upload_bytes_per_second": 4910712.282556662,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.84579917162814,
+    "estimated_full_shards_per_hour": 85.82362008245119,
     "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": 1.0245182449980348,
-      "bytes": 6075842,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3879416819981998,
       "bytes": 11260844,
@@ -562,6 +556,12 @@
       "bytes": 2387118,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1524969450001663,
+      "bytes": 6155701,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 421600, "loss": 0.5581619739532471, "lr": 0.0003, "grad_norm": 0.15593133275044313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61710802049, "tokens_per_second": 149947.38246147762, "data_wait_s": 8.893999620340765e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 421700, "loss": 0.5913926959037781, "lr": 0.0003, "grad_norm": 0.17915203021932297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61725345966, "tokens_per_second": 151287.51366720232, "data_wait_s": 8.082999556791037e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 421800, "loss": 0.6039777398109436, "lr": 0.0003, "grad_norm": 0.1980378037377272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61740025221, "tokens_per_second": 151331.86318145675, "data_wait_s": 1.2757001968566328e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 421900, "loss": 0.6058818697929382, "lr": 0.0003, "grad_norm": 0.1551188317121638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61754736801, "tokens_per_second": 152753.96167593866, "data_wait_s": 1.0063999070553109e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f75e4c6587bf3cf01677488bd1621a705afa6a986a43decb3edcc94238d994ed
-size 527226
+oid sha256:2163677f3453ee629232ccbc8abf95267f8ff60761c9b6b63b47dbac4f09cd5c
+size 529344