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": 75969.84804358913,
-      "required_pages_per_second_upper_estimate": 0.005255770361988821
+      "required_token_positions_per_second": 76339.77126097363,
+      "required_pages_per_second_upper_estimate": 0.005281362508507601
     },
     {
       "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": 9496.23100544864,
-      "required_pages_per_second_upper_estimate": 0.003975873618128069
+      "required_token_positions_per_second": 9542.471407621704,
+      "required_pages_per_second_upper_estimate": 0.003995233508908531
     },
     {
       "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": 18992.46201089728,
-      "required_pages_per_second_upper_estimate": 0.00395562897196497
+      "required_token_positions_per_second": 19084.94281524341,
+      "required_pages_per_second_upper_estimate": 0.00397489028462745
     },
     {
       "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": 28488.693016345926,
-      "required_pages_per_second_upper_estimate": 0.0017078536466144175
+      "required_token_positions_per_second": 28627.414222865114,
+      "required_pages_per_second_upper_estimate": 0.001716169771130225
     },
     {
       "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": 28488.693016345926,
-      "required_pages_per_second_upper_estimate": 0.0022148705735033293
+      "required_token_positions_per_second": 28627.414222865114,
+      "required_pages_per_second_upper_estimate": 0.0022256555371400932
     },
     {
       "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": 28488.693016345926,
-      "required_pages_per_second_upper_estimate": 0.36542236523833616
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28627.414222865114,
+      "required_pages_per_second_upper_estimate": 0.36720173192833744
     }
   ],
   "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": 0.08369665069537291,
-    "producer_p95_seconds": 0.24592177299928153,
-    "gpu_update_seconds": 0.9704902918549629,
-    "producer_batches_per_second": 11.947909404877585,
-    "gpu_batches_per_second": 1.0304070101398266,
-    "producer_headroom": 11.595330085396304,
+    "producer_batch_seconds": 0.09313316457817677,
+    "producer_p95_seconds": 0.21246552699994936,
+    "gpu_update_seconds": 0.9657875414370173,
+    "producer_batches_per_second": 10.737313657591775,
+    "gpu_batches_per_second": 1.035424414889508,
+    "producer_headroom": 10.369963759003667,
     "sustainable": true,
-    "buffer_seconds": 60.170398095007705,
+    "buffer_seconds": 60.84461511053209,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394048.47207081615,
-    "upload_bytes_per_second": 4014742.565495894,
+    "generation_bytes_per_second": 393499.93090226204,
+    "upload_bytes_per_second": 3924430.2119644955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.55362912743915,
+    "estimated_full_shards_per_hour": 84.43592496205231,
     "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": 1.5575974499997756,
-      "bytes": 11152630,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.9014979050007241,
       "bytes": 5893171,
@@ -562,6 +556,12 @@
       "bytes": 6201366,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4180366079999658,
+      "bytes": 818969,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -124,3 +124,4 @@
 {"step": 812400, "loss": 0.5304609537124634, "lr": 3e-05, "grad_norm": 0.2786341261923443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118912177957, "tokens_per_second": 154063.14123069774, "data_wait_s": 1.5179999536485411e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812500, "loss": 0.53594571352005, "lr": 3e-05, "grad_norm": 0.2830846785903361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118926928546, "tokens_per_second": 151748.15153972426, "data_wait_s": 1.848800093284808e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812600, "loss": 0.5150604844093323, "lr": 3e-05, "grad_norm": 0.4109485723270542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118941677595, "tokens_per_second": 152932.23902563492, "data_wait_s": 9.286999556934461e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 812700, "loss": 0.509462296962738, "lr": 3e-05, "grad_norm": 0.26544499288587536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118956098928, "tokens_per_second": 149816.37113361756, "data_wait_s": 1.434399928257335e-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 sha256:23e3bcb4bfea203bede199cc5b31f259a7a9e3e9edc033dbfabcdf4c87f3c2f1
-size 268599
+oid sha256:37a6a7ca99f6e195468993e22821bdd8e896c633ee2300e73c9029435cc5caf6
+size 270717