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": 75859.94326902524,
-      "required_pages_per_second_upper_estimate": 0.00524816689467028
+      "required_token_positions_per_second": 76452.01994847241,
+      "required_pages_per_second_upper_estimate": 0.005289128133161606
     },
     {
       "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": 9482.492908628155,
-      "required_pages_per_second_upper_estimate": 0.003970121763873416
+      "required_token_positions_per_second": 9556.502493559052,
+      "required_pages_per_second_upper_estimate": 0.004001108031588105
     },
     {
       "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": 18964.98581725631,
-      "required_pages_per_second_upper_estimate": 0.0039499064054255096
+      "required_token_positions_per_second": 19113.004987118104,
+      "required_pages_per_second_upper_estimate": 0.003980734894979711
     },
     {
       "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": 28447.478725884463,
-      "required_pages_per_second_upper_estimate": 0.0017053829128318306
+      "required_token_positions_per_second": 28669.507480677155,
+      "required_pages_per_second_upper_estimate": 0.0017186931976633736
     },
     {
       "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": 28447.478725884463,
-      "required_pages_per_second_upper_estimate": 0.0022116663437024555
+      "required_token_positions_per_second": 28669.507480677155,
+      "required_pages_per_second_upper_estimate": 0.0022289281027863058
     },
     {
       "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": 28447.478725884463,
-      "required_pages_per_second_upper_estimate": 0.3648937125727538
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28669.507480677155,
+      "required_pages_per_second_upper_estimate": 0.36774165904333134
     }
   ],
   "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": 0.10483702262501993,
-    "producer_p95_seconds": 0.2908202760008862,
-    "gpu_update_seconds": 0.9718963239734488,
-    "producer_batches_per_second": 9.538615032752222,
-    "gpu_batches_per_second": 1.0289163312313536,
-    "producer_headroom": 9.270544886129764,
+    "producer_batch_seconds": 0.3321738296640433,
+    "producer_p95_seconds": 0.29735259199878783,
+    "gpu_update_seconds": 0.964369549028157,
+    "producer_batches_per_second": 3.010471959851227,
+    "gpu_batches_per_second": 1.0369468851518069,
+    "producer_headroom": 2.9032074862836397,
     "sustainable": true,
-    "buffer_seconds": 60.257572086353825,
+    "buffer_seconds": 60.75528158877389,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307883.6551442148,
-    "upload_bytes_per_second": 4793516.517513572,
+    "generation_bytes_per_second": 337523.4591588859,
+    "upload_bytes_per_second": 4773346.811410508,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.06466522927126,
+    "estimated_full_shards_per_hour": 72.42467719149525,
     "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": 1.176057447999483,
-      "bytes": 7472528,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.2658942909984034,
       "bytes": 6109615,
@@ -562,6 +556,12 @@
       "bytes": 7122874,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2547058620002645,
+      "bytes": 6059042,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 790900, "loss": 0.5075185894966125, "lr": 3e-05, "grad_norm": 0.25775752807402286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115765132292, "tokens_per_second": 151411.71402798707, "data_wait_s": 8.085000445134938e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 791000, "loss": 0.507858395576477, "lr": 3e-05, "grad_norm": 0.2734002704141276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115779772392, "tokens_per_second": 152049.74784515027, "data_wait_s": 1.0251002095174044e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791100, "loss": 0.4920814335346222, "lr": 3e-05, "grad_norm": 0.2595631820435781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115794328226, "tokens_per_second": 152004.21643786854, "data_wait_s": 1.529400105937384e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 791200, "loss": 0.5363965630531311, "lr": 3e-05, "grad_norm": 0.278750405515782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115808961492, "tokens_per_second": 151780.25353076376, "data_wait_s": 1.0931002179859206e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5af1fadc56c8a4dbf1451c771e6c48a112b4732fa8e7820e4c3ec629d53e3bf
-size 663561
+oid sha256:6e2a900181d79e69bc4cd78b6404ceb4d8ddf7ab4125e3aefdc432696a82123e
+size 665679