Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75453.75028647632,
-      "required_pages_per_second_upper_estimate": 0.0065401398390500295
+      "required_token_positions_per_second": 73111.38662005353,
+      "required_pages_per_second_upper_estimate": 0.006337109693110938
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9431.71878580954,
-      "required_pages_per_second_upper_estimate": 0.0037624536404218686
+      "required_token_positions_per_second": 9138.92332750669,
+      "required_pages_per_second_upper_estimate": 0.0036456531544226465
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18863.43757161908,
-      "required_pages_per_second_upper_estimate": 0.004363724297031515
+      "required_token_positions_per_second": 18277.84665501338,
+      "required_pages_per_second_upper_estimate": 0.0042282581445229235
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28295.15635742862,
-      "required_pages_per_second_upper_estimate": 0.0017230265689550642
+      "required_token_positions_per_second": 27416.769982520076,
+      "required_pages_per_second_upper_estimate": 0.0016695374472602766
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28295.15635742862,
-      "required_pages_per_second_upper_estimate": 0.00217823946240074
+      "required_token_positions_per_second": 27416.769982520076,
+      "required_pages_per_second_upper_estimate": 0.00211061884773117
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28295.15635742862,
-      "required_pages_per_second_upper_estimate": 0.35133114416266586
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27416.769982520076,
+      "required_pages_per_second_upper_estimate": 0.3404245251563887
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.448939788249902,
-    "producer_p95_seconds": 0.41167778599628946,
-    "gpu_update_seconds": 0.9771283696314081,
-    "producer_batches_per_second": 2.227470200176044,
-    "gpu_batches_per_second": 1.0234069863074586,
-    "producer_headroom": 2.1765243251005644,
+    "producer_batch_seconds": 0.25408716698439093,
+    "producer_p95_seconds": 0.4104651669986197,
+    "gpu_update_seconds": 1.008433889828282,
+    "producer_batches_per_second": 3.9356572465599253,
+    "gpu_batches_per_second": 0.9916366457798059,
+    "producer_headroom": 3.968850146179292,
     "sustainable": true,
-    "buffer_seconds": 60.581958917147304,
+    "buffer_seconds": 60.506033389696924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390117.98825014813,
-    "upload_bytes_per_second": 3684364.3306309036,
+    "generation_bytes_per_second": 369926.34108565963,
+    "upload_bytes_per_second": 3757561.1149034123,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.71023879650434,
+    "estimated_full_shards_per_hour": 79.37758135249464,
     "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": 0.9772542809987499,
-      "bytes": 2642702,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8478547829981835,
       "bytes": 5863488,
@@ -562,6 +556,12 @@
       "bytes": 1072664,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.132626735001395,
+      "bytes": 11459123,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 38700, "loss": 0.6552939936518669, "lr": 0.0003, "grad_norm": 0.12835237871211294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5663203830, "tokens_per_second": 149862.03059944644, "data_wait_s": 9.670999133959413e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 38800, "loss": 0.6750294882804155, "lr": 0.0003, "grad_norm": 0.15139465182627893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5677886427, "tokens_per_second": 148717.4995696666, "data_wait_s": 1.4072000340092927e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38900, "loss": 0.6260733418166637, "lr": 0.0003, "grad_norm": 0.12423666107829527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5692507295, "tokens_per_second": 148968.90412379554, "data_wait_s": 1.243100268766284e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 39000, "loss": 0.645189106464386, "lr": 0.0003, "grad_norm": 0.13044886909258227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5706920096, "tokens_per_second": 145370.7605894049, "data_wait_s": 7.5930001912638545e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:688834175514699a9ab40d7de9aee701c78c98b4169c1b5a3be30db51b1d04aa
-size 823774
+oid sha256:adc0d0bbeb8f007fe377c0fc76ebd450cae382984bf6c47511fc436eb753b2f4
+size 825892