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": 76061.41339155668,
-      "required_pages_per_second_upper_estimate": 0.005262105065222095
+      "required_token_positions_per_second": 76223.80083440965,
+      "required_pages_per_second_upper_estimate": 0.0052733394053094415
     },
     {
       "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": 9507.676673944585,
-      "required_pages_per_second_upper_estimate": 0.0039806656805145835
+      "required_token_positions_per_second": 9527.975104301206,
+      "required_pages_per_second_upper_estimate": 0.003989164209425474
     },
     {
       "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": 19015.35334788917,
-      "required_pages_per_second_upper_estimate": 0.003960396633775228
+      "required_token_positions_per_second": 19055.95020860241,
+      "required_pages_per_second_upper_estimate": 0.0039688518892505836
     },
     {
       "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": 28523.030021833754,
-      "required_pages_per_second_upper_estimate": 0.0017099120976638413
+      "required_token_positions_per_second": 28583.925312903615,
+      "required_pages_per_second_upper_estimate": 0.0017135626773817538
     },
     {
       "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": 28523.030021833754,
-      "required_pages_per_second_upper_estimate": 0.0022175401246474825
+      "required_token_positions_per_second": 28583.925312903615,
+      "required_pages_per_second_upper_estimate": 0.0022222744656780857
     },
     {
       "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": 28523.030021833754,
-      "required_pages_per_second_upper_estimate": 0.36586280347653
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28583.925312903615,
+      "required_pages_per_second_upper_estimate": 0.36664390288610477
     }
   ],
   "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": 0.4888679152577424,
-    "producer_p95_seconds": 0.27726509899730445,
-    "gpu_update_seconds": 0.9693219822310625,
-    "producer_batches_per_second": 2.0455423004653044,
-    "gpu_batches_per_second": 1.0316489446554453,
-    "producer_headroom": 1.9827891174245167,
+    "producer_batch_seconds": 0.1265654133438261,
+    "producer_p95_seconds": 0.27417943999898853,
+    "gpu_update_seconds": 0.9672569354048406,
+    "producer_batches_per_second": 7.90105269346699,
+    "gpu_batches_per_second": 1.0338514653104607,
+    "producer_headroom": 7.642348014755042,
     "sustainable": true,
-    "buffer_seconds": 60.097962898325875,
+    "buffer_seconds": 60.93718693050496,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418138.05875493854,
-    "upload_bytes_per_second": 5222746.663379439,
+    "generation_bytes_per_second": 418693.9806341903,
+    "upload_bytes_per_second": 5319065.246347881,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.72269365297429,
+    "estimated_full_shards_per_hour": 89.8419815470627,
     "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": 1.3663103399994725,
-      "bytes": 5168575,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.1543622389999655,
       "bytes": 6027007,
@@ -562,6 +556,12 @@
       "bytes": 6115368,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3705160650024482,
+      "bytes": 13033658,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 743000, "loss": 0.5352374315261841, "lr": 5e-05, "grad_norm": 0.22055568912012133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108754935247, "tokens_per_second": 148846.01957060315, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743100, "loss": 0.5451949238777161, "lr": 5e-05, "grad_norm": 0.2491799781659081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108769514755, "tokens_per_second": 150268.03400213382, "data_wait_s": 1.3527998817153275e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743200, "loss": 0.5074306130409241, "lr": 5e-05, "grad_norm": 0.22568622470124966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108784215551, "tokens_per_second": 151115.52954239104, "data_wait_s": 1.3770000805379823e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 743300, "loss": 0.571385383605957, "lr": 5e-05, "grad_norm": 0.2381628199004266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108798847193, "tokens_per_second": 150964.88080803826, "data_wait_s": 1.134899866883643e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba662c8bfec4a61556319e7fc113d9e58ae87cce23de0e6f630fbc50b5b822dd
-size 494673
+oid sha256:cada70f8ed097083478b54d6a30ddcf9a980f6d2aff7bf9181fda48aceabed79
+size 496791