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": 75669.95432538974,
-      "required_pages_per_second_upper_estimate": 0.005235023018714492
+      "required_token_positions_per_second": 73866.07389043708,
+      "required_pages_per_second_upper_estimate": 0.0051102263846451975
     },
     {
       "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": 9458.744290673718,
-      "required_pages_per_second_upper_estimate": 0.00396017871346343
+      "required_token_positions_per_second": 9233.259236304635,
+      "required_pages_per_second_upper_estimate": 0.0038657728298624704
     },
     {
       "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": 18917.488581347436,
-      "required_pages_per_second_upper_estimate": 0.003940013983772034
+      "required_token_positions_per_second": 18466.51847260927,
+      "required_pages_per_second_upper_estimate": 0.0038460888030034035
     },
     {
       "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": 28376.232872021155,
-      "required_pages_per_second_upper_estimate": 0.0017011118326788453
+      "required_token_positions_per_second": 27699.77770891391,
+      "required_pages_per_second_upper_estimate": 0.0016605593785377938
     },
     {
       "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": 28376.232872021155,
-      "required_pages_per_second_upper_estimate": 0.002206127291941976
+      "required_token_positions_per_second": 27699.77770891391,
+      "required_pages_per_second_upper_estimate": 0.002153535878422199
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28376.232872021155,
-      "required_pages_per_second_upper_estimate": 0.36397984725723315
+      "required_token_positions_per_second": 27699.77770891391,
+      "required_pages_per_second_upper_estimate": 0.35530300674585896
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11884813185139365,
+    "producer_batch_seconds": 0.5788673366012063,
     "producer_p95_seconds": 0.35542541600443656,
-    "gpu_update_seconds": 0.974336520449859,
-    "producer_batches_per_second": 8.41409944289565,
-    "gpu_batches_per_second": 1.0263394412623392,
-    "producer_headroom": 8.198164373910044,
+    "gpu_update_seconds": 0.9981307536306602,
+    "producer_batches_per_second": 1.7275115329039903,
+    "gpu_batches_per_second": 1.0018727469948607,
+    "producer_headroom": 1.7242823882431169,
     "sustainable": true,
-    "buffer_seconds": 60.40886426789125,
+    "buffer_seconds": 60.88597597147027,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405178.5552262043,
-    "upload_bytes_per_second": 4567530.874202559,
+    "generation_bytes_per_second": 375998.3905389349,
+    "upload_bytes_per_second": 4574610.518601385,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.94188587751003,
+    "estimated_full_shards_per_hour": 80.68050181509052,
     "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": 1.4454559629994037,
-      "bytes": 6189473,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.604715366000164,
       "bytes": 6773206,
@@ -562,6 +556,12 @@
       "bytes": 7488941,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.3952881760051241,
+      "bytes": 6629850,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 644500, "loss": 0.564680278301239, "lr": 0.0001, "grad_norm": 0.19166007721770675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94338417294, "tokens_per_second": 152386.79696061014, "data_wait_s": 1.418799365637824e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644600, "loss": 0.5822890400886536, "lr": 0.0001, "grad_norm": 0.22338748039960604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94352959361, "tokens_per_second": 150792.3327680029, "data_wait_s": 1.3247001334093511e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644700, "loss": 0.5330870151519775, "lr": 0.0001, "grad_norm": 0.1697825474437093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94367436580, "tokens_per_second": 149924.6990900405, "data_wait_s": 1.9946004613302648e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 644800, "loss": 0.5745224952697754, "lr": 0.0001, "grad_norm": 0.17889248196152127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94381940687, "tokens_per_second": 144931.13760888163, "data_wait_s": 7.866998203098774e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d41b1cc45719b2eabe340825dc9b93b41fe4ca3a91742b5b343b0c972e7573ff
-size 859983
+oid sha256:b9ae22d8c3d1397fc5e3e9033276d7e7cfd4271b12038c554df0f6b4696caa06
+size 862101