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": 76076.83151629835,
-      "required_pages_per_second_upper_estimate": 0.005263171726866702
+      "required_token_positions_per_second": 77037.86649548938,
+      "required_pages_per_second_upper_estimate": 0.0053296583566355095
     },
     {
       "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": 9509.603939537294,
-      "required_pages_per_second_upper_estimate": 0.003981472586372369
+      "required_token_positions_per_second": 9629.733311936172,
+      "required_pages_per_second_upper_estimate": 0.004031768246009222
     },
     {
       "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": 19019.207879074587,
-      "required_pages_per_second_upper_estimate": 0.0039611994309703275
+      "required_token_positions_per_second": 19259.466623872344,
+      "required_pages_per_second_upper_estimate": 0.004011238991462519
     },
     {
       "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": 28528.811818611885,
-      "required_pages_per_second_upper_estimate": 0.0017102587075523976
+      "required_token_positions_per_second": 28889.199935808516,
+      "required_pages_per_second_upper_estimate": 0.001731863424897543
     },
     {
       "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": 28528.811818611885,
-      "required_pages_per_second_upper_estimate": 0.0022179896339155456
+      "required_token_positions_per_second": 28889.199935808516,
+      "required_pages_per_second_upper_estimate": 0.002246008224847783
     },
     {
       "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": 28528.811818611885,
-      "required_pages_per_second_upper_estimate": 0.36593696615759014
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28889.199935808516,
+      "required_pages_per_second_upper_estimate": 0.37055963797037644
     }
   ],
   "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": 0.07543604178120233,
-    "producer_p95_seconds": 0.21802650700010417,
-    "gpu_update_seconds": 0.969125534417201,
-    "producer_batches_per_second": 13.256262873659773,
-    "gpu_batches_per_second": 1.0318580663560433,
-    "producer_headroom": 12.846982841810428,
+    "producer_batch_seconds": 0.07563107155469595,
+    "producer_p95_seconds": 0.21379042199987452,
+    "gpu_update_seconds": 0.9570358494327829,
+    "producer_batches_per_second": 13.222078960983724,
+    "gpu_batches_per_second": 1.0448929374930742,
+    "producer_headroom": 12.654003569692387,
     "sustainable": true,
-    "buffer_seconds": 60.08578313386646,
+    "buffer_seconds": 60.29325851426532,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 514583.7902022009,
-    "upload_bytes_per_second": 3966362.247851994,
+    "generation_bytes_per_second": 514934.53732749156,
+    "upload_bytes_per_second": 3990288.2977678296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.41770247983474,
+    "estimated_full_shards_per_hour": 110.49296464794693,
     "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": 1.460268009000174,
-      "bytes": 1130080,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.6416546800001015,
       "bytes": 7689453,
@@ -562,6 +556,12 @@
       "bytes": 8470485,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.9930366190001223,
+      "bytes": 5866239,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -136,3 +136,4 @@
 {"step": 813600, "loss": 0.620110034942627, "lr": 3e-05, "grad_norm": 0.26412038073598465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119087648976, "tokens_per_second": 150222.12689090133, "data_wait_s": 6.595999366254546e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 813700, "loss": 0.5054027438163757, "lr": 3e-05, "grad_norm": 0.2698917196061391, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119102279338, "tokens_per_second": 150872.49377637048, "data_wait_s": 1.0710999049479142e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813800, "loss": 0.5414049029350281, "lr": 3e-05, "grad_norm": 0.2815565741459654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119116976455, "tokens_per_second": 152972.4789200797, "data_wait_s": 1.6899000911507756e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 813900, "loss": 0.529843270778656, "lr": 3e-05, "grad_norm": 0.2860852396890787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119131526765, "tokens_per_second": 150183.54668206448, "data_wait_s": 1.9241000700276345e-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 sha256:848d27c6ca750200005aaaec9729722b4191201578d22644144609aa1ed49a3f
-size 294015
+oid sha256:cc05ab41f044bb058d1a299f3cd905bf20485600d56269de587758339552d190
+size 296133