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": 77200.14812133243,
-      "required_pages_per_second_upper_estimate": 0.005340885376056582
+      "required_token_positions_per_second": 76905.61142976853,
+      "required_pages_per_second_upper_estimate": 0.00532050864431491
     },
     {
       "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": 9650.018515166554,
-      "required_pages_per_second_upper_estimate": 0.004040261236998571
+      "required_token_positions_per_second": 9613.201428721066,
+      "required_pages_per_second_upper_estimate": 0.004024846691732036
     },
     {
       "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": 19300.037030333107,
-      "required_pages_per_second_upper_estimate": 0.004019688737214729
+      "required_token_positions_per_second": 19226.40285744213,
+      "required_pages_per_second_upper_estimate": 0.004004352680865263
     },
     {
       "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": 28950.055545499657,
-      "required_pages_per_second_upper_estimate": 0.0017355116257773912
+      "required_token_positions_per_second": 28839.604286163198,
+      "required_pages_per_second_upper_estimate": 0.0017288902414294785
     },
     {
       "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": 28950.055545499657,
-      "required_pages_per_second_upper_estimate": 0.002250739480825731
+      "required_token_positions_per_second": 28839.604286163198,
+      "required_pages_per_second_upper_estimate": 0.0022421523812360677
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28950.055545499657,
-      "required_pages_per_second_upper_estimate": 0.37134022838983155
+      "required_token_positions_per_second": 28839.604286163198,
+      "required_pages_per_second_upper_estimate": 0.36992347822838595
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.372523218593642,
-    "producer_p95_seconds": 0.18958198000109405,
-    "gpu_update_seconds": 0.9550240743596063,
-    "producer_batches_per_second": 2.6843964351409353,
-    "gpu_batches_per_second": 1.0470940229130374,
-    "producer_headroom": 2.563663220684699,
+    "producer_batch_seconds": 0.07567663154674165,
+    "producer_p95_seconds": 0.18841723799960164,
+    "gpu_update_seconds": 0.9586816700278058,
+    "producer_batches_per_second": 13.21411880472442,
+    "gpu_batches_per_second": 1.0430991133594907,
+    "producer_headroom": 12.66813348365904,
     "sustainable": true,
-    "buffer_seconds": 60.1665166846552,
+    "buffer_seconds": 60.39694521175176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305467.3644918682,
-    "upload_bytes_per_second": 4584941.787631983,
+    "generation_bytes_per_second": 305062.6216008596,
+    "upload_bytes_per_second": 4496015.548259048,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.54618550364526,
+    "estimated_full_shards_per_hour": 65.45933710116712,
     "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": 1.2533257390000472,
-      "bytes": 16191062,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.8292768119999891,
       "bytes": 2252671,
@@ -562,6 +556,12 @@
       "bytes": 12237678,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.9231663379996462,
+      "bytes": 6121395,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 448400, "loss": 0.5554867386817932, "lr": 0.0003, "grad_norm": 0.14706922969069713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65631727393, "tokens_per_second": 151249.41875372521, "data_wait_s": 1.0630999895511195e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448500, "loss": 0.5858342051506042, "lr": 0.0003, "grad_norm": 0.15514609296353427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65646337332, "tokens_per_second": 152351.41002405842, "data_wait_s": 9.543000487610698e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 448600, "loss": 0.5526136159896851, "lr": 0.0003, "grad_norm": 0.18028248759067692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65661268365, "tokens_per_second": 155077.3128269652, "data_wait_s": 1.3885999578633346e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 448700, "loss": 0.5940396189689636, "lr": 0.0003, "grad_norm": 0.15803821648863708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65676020290, "tokens_per_second": 152803.27240567782, "data_wait_s": 1.5137999071157537e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68a6331e133dcea3be555f5aafb95a336ba3b9191651cb2d0d08d9cd717965f6
-size 178860
+oid sha256:edbe91e3e1702c8a550f39e4ebff1ed35f64a7f7b9d24aaaee08984f71197409
+size 180978