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": 76856.80892439833,
-      "required_pages_per_second_upper_estimate": 0.005317132373755986
+      "required_token_positions_per_second": 75572.1882942542,
+      "required_pages_per_second_upper_estimate": 0.0052282593378320886
     },
     {
       "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": 9607.10111554979,
-      "required_pages_per_second_upper_estimate": 0.004022292617996248
+      "required_token_positions_per_second": 9446.523536781775,
+      "required_pages_per_second_upper_estimate": 0.003955062139007234
     },
     {
       "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": 19214.20223109958,
-      "required_pages_per_second_upper_estimate": 0.004001811612150263
+      "required_token_positions_per_second": 18893.04707356355,
+      "required_pages_per_second_upper_estimate": 0.0039349234622665565
     },
     {
       "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": 28821.303346649373,
-      "required_pages_per_second_upper_estimate": 0.0017277931280495413
+      "required_token_positions_per_second": 28339.570610345327,
+      "required_pages_per_second_upper_estimate": 0.0016989139860714114
     },
     {
       "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": 28821.303346649373,
-      "required_pages_per_second_upper_estimate": 0.0022407295636861935
+      "required_token_positions_per_second": 28339.570610345327,
+      "required_pages_per_second_upper_estimate": 0.002203276962356927
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28821.303346649373,
-      "required_pages_per_second_upper_estimate": 0.3696887334263205
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28339.570610345327,
+      "required_pages_per_second_upper_estimate": 0.36350958312932524
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10652082011748121,
-    "producer_p95_seconds": 0.27853430200048024,
-    "gpu_update_seconds": 0.9592904133259548,
-    "producer_batches_per_second": 9.387836095301422,
-    "gpu_batches_per_second": 1.0424371870171214,
-    "producer_headroom": 9.005661168098019,
+    "producer_batch_seconds": 0.40485083310943537,
+    "producer_p95_seconds": 0.3002608570022858,
+    "gpu_update_seconds": 0.9755969975743786,
+    "producer_batches_per_second": 2.4700455531227465,
+    "gpu_batches_per_second": 1.025013404598717,
+    "producer_headroom": 2.4097690254984965,
     "sustainable": true,
-    "buffer_seconds": 60.43529603953515,
+    "buffer_seconds": 60.48701384961147,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317562.19182567426,
-    "upload_bytes_per_second": 3238230.224910751,
+    "generation_bytes_per_second": 294549.1564917664,
+    "upload_bytes_per_second": 3206208.7409059615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.14145389630957,
+    "estimated_full_shards_per_hour": 63.20339222969765,
     "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": 1.7584444239982986,
-      "bytes": 6116166,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7439238019978802,
       "bytes": 6171251,
@@ -562,6 +556,12 @@
       "bytes": 11961950,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 1.958790611999575,
+      "bytes": 2439502,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 829400, "loss": 0.5459651947021484, "lr": 3e-05, "grad_norm": 0.28340184518973255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121401004413, "tokens_per_second": 153089.67534885972, "data_wait_s": 8.663999324198812e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 829500, "loss": 0.5509658455848694, "lr": 3e-05, "grad_norm": 0.2958763041286958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121415610041, "tokens_per_second": 151747.01386022512, "data_wait_s": 1.4760000340174884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829600, "loss": 0.5377776026725769, "lr": 3e-05, "grad_norm": 0.2788177400631082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121430194259, "tokens_per_second": 150181.48819472408, "data_wait_s": 1.1932999768760055e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 829700, "loss": 0.5087186694145203, "lr": 3e-05, "grad_norm": 0.3078371649183931, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121444782976, "tokens_per_second": 151170.97979522677, "data_wait_s": 1.2895998224848881e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebd67d219b12ac5f14064e5cce0f9edf988faaeef378de43f782990063ea8de4
-size 631008
+oid sha256:cd5e14de6192b5fbd4fc673ac22320191ec02fc36f3affbd82c8eb994a301286
+size 633126