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": 75119.61204561256,
-      "required_pages_per_second_upper_estimate": 0.005196949062829485
+      "required_token_positions_per_second": 77078.1231222485,
+      "required_pages_per_second_upper_estimate": 0.0053324434034829485
     },
     {
       "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": 9389.95150570157,
-      "required_pages_per_second_upper_estimate": 0.003931376610952299
+      "required_token_positions_per_second": 9634.765390281062,
+      "required_pages_per_second_upper_estimate": 0.004033875072130483
     },
     {
       "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": 18779.90301140314,
-      "required_pages_per_second_upper_estimate": 0.0039113585379281154
+      "required_token_positions_per_second": 19269.530780562123,
+      "required_pages_per_second_upper_estimate": 0.00401333508989134
     },
     {
       "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": 28169.85451710471,
-      "required_pages_per_second_upper_estimate": 0.0016887397654231063
+      "required_token_positions_per_second": 28904.296170843183,
+      "required_pages_per_second_upper_estimate": 0.001732768421137255
     },
     {
       "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": 28169.85451710471,
-      "required_pages_per_second_upper_estimate": 0.0021900822826096704
+      "required_token_positions_per_second": 28904.296170843183,
+      "required_pages_per_second_upper_estimate": 0.002247181890720408
     },
     {
       "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": 28169.85451710471,
-      "required_pages_per_second_upper_estimate": 0.361332647312178
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28904.296170843183,
+      "required_pages_per_second_upper_estimate": 0.3707532762643268
     }
   ],
   "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": 0.08707207583586296,
-    "producer_p95_seconds": 0.3059780490002595,
-    "gpu_update_seconds": 0.9814747173512081,
-    "producer_batches_per_second": 11.484738251618934,
-    "gpu_batches_per_second": 1.0188749463651878,
-    "producer_headroom": 11.2719802293603,
+    "producer_batch_seconds": 0.25313794990682936,
+    "producer_p95_seconds": 0.33374022800126113,
+    "gpu_update_seconds": 0.9565360054637672,
+    "producer_batches_per_second": 3.95041518021325,
+    "gpu_batches_per_second": 1.0454389529384833,
+    "producer_headroom": 3.77871435640461,
     "sustainable": true,
-    "buffer_seconds": 60.8514324757749,
+    "buffer_seconds": 60.26176834421733,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404580.0249721051,
-    "upload_bytes_per_second": 5427032.875671282,
+    "generation_bytes_per_second": 399680.4715005199,
+    "upload_bytes_per_second": 5332687.183243483,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.81345521805157,
+    "estimated_full_shards_per_hour": 85.76212509881685,
     "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": 1.1704257509991294,
-      "bytes": 15475495,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.1073925720011175,
       "bytes": 2592720,
@@ -562,6 +556,12 @@
       "bytes": 5725754,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0718349349990604,
+      "bytes": 7504941,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -343,3 +343,4 @@
 {"step": 474500, "loss": 0.5762530565261841, "lr": 0.0003, "grad_norm": 0.17740558733894285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69451916420, "tokens_per_second": 152573.086350674, "data_wait_s": 1.3256001693662256e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 474600, "loss": 0.5983161330223083, "lr": 0.0003, "grad_norm": 0.1665082044412824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69466474121, "tokens_per_second": 151087.0543587786, "data_wait_s": 1.1262003681622446e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 474700, "loss": 0.5879476070404053, "lr": 0.0003, "grad_norm": 0.14858122517042577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69480994923, "tokens_per_second": 150798.3007386908, "data_wait_s": 9.6789954113774e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 474800, "loss": 0.5864902138710022, "lr": 0.0003, "grad_norm": 0.15339896740445758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69495542232, "tokens_per_second": 151831.1309800085, "data_wait_s": 9.854004019871354e-06, "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 sha256:645645e1ac322b62f6147cc35a9ce7eb20ac83efc3b133aa7fc9df7da3e265a4
-size 735573
+oid sha256:974cc50d0771b8448afdb83b73ef66825e3ceff823c1f0bfccb47e545d5a9a82
+size 737691