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": 76479.8037306057,
-      "required_pages_per_second_upper_estimate": 0.005291050279676842
+      "required_token_positions_per_second": 76220.86122565938,
+      "required_pages_per_second_upper_estimate": 0.005273136036355266
     },
     {
       "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": 9559.975466325712,
-      "required_pages_per_second_upper_estimate": 0.004002562092761587
+      "required_token_positions_per_second": 9527.607653207422,
+      "required_pages_per_second_upper_estimate": 0.003989010365325759
     },
     {
       "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": 19119.950932651423,
-      "required_pages_per_second_upper_estimate": 0.003982181552257399
+      "required_token_positions_per_second": 19055.215306414844,
+      "required_pages_per_second_upper_estimate": 0.003968698828505588
     },
     {
       "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": 28679.926398977135,
-      "required_pages_per_second_upper_estimate": 0.0017193177959066903
+      "required_token_positions_per_second": 28582.822959622266,
+      "required_pages_per_second_upper_estimate": 0.001713496592985733
     },
     {
       "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": 28679.926398977135,
-      "required_pages_per_second_upper_estimate": 0.002229738127856151
+      "required_token_positions_per_second": 28582.822959622266,
+      "required_pages_per_second_upper_estimate": 0.0022221887625592006
     },
     {
       "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": 28679.926398977135,
-      "required_pages_per_second_upper_estimate": 0.3678753017403206
+      "required_token_positions_per_second": 28582.822959622266,
+      "required_pages_per_second_upper_estimate": 0.3666297630818264
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.33688159614058577,
-    "producer_p95_seconds": 0.23325053499866044,
-    "gpu_update_seconds": 0.964019210348673,
-    "producer_batches_per_second": 2.9684019888776736,
-    "gpu_batches_per_second": 1.0373237268148559,
-    "producer_headroom": 2.861596541315285,
+    "producer_batch_seconds": 0.07062963224231567,
+    "producer_p95_seconds": 0.25068659099997603,
+    "gpu_update_seconds": 0.9672942395877814,
+    "producer_batches_per_second": 14.158363398654076,
+    "gpu_batches_per_second": 1.0338115943150414,
+    "producer_headroom": 13.695303357508571,
     "sustainable": true,
-    "buffer_seconds": 60.7332102519664,
+    "buffer_seconds": 60.93953709403023,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319100.3770606894,
-    "upload_bytes_per_second": 1917020.5118191184,
+    "generation_bytes_per_second": 337575.66775533656,
+    "upload_bytes_per_second": 1916722.391458404,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.47151264062416,
+    "estimated_full_shards_per_hour": 72.43587994093964,
     "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": 1.1174767350003094,
-      "bytes": 6113163,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3073732009997912,
       "bytes": 7581885,
@@ -562,6 +556,12 @@
       "bytes": 5705697,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1611206180023146,
+      "bytes": 6127791,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 664400, "loss": 0.5364418625831604, "lr": 0.0001, "grad_norm": 0.1788650947893196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97251111143, "tokens_per_second": 153055.23395418428, "data_wait_s": 9.680003131506965e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 664500, "loss": 0.5686577558517456, "lr": 0.0001, "grad_norm": 0.19050801079009147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97265774477, "tokens_per_second": 152066.24231930618, "data_wait_s": 9.14699921850115e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 664600, "loss": 0.5870509743690491, "lr": 0.0001, "grad_norm": 0.19711691401713738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97280553407, "tokens_per_second": 153991.92719648144, "data_wait_s": 9.594001312507316e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 664700, "loss": 0.5454566478729248, "lr": 0.0001, "grad_norm": 0.19392334602321665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97295201154, "tokens_per_second": 152737.8006900432, "data_wait_s": 9.757997759152204e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16e46935682473b25e8927ab17296f4d77c0d0a014c63ebd7d9120589df13cfd
-size 417642
+oid sha256:caf03ae63cc6c8369508a4a1a41302c83db419e13124727438c4544e3eb68323
+size 419760