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": 75582.41639617617,
-      "required_pages_per_second_upper_estimate": 0.006551292291337538
+      "required_token_positions_per_second": 73754.44326234529,
+      "required_pages_per_second_upper_estimate": 0.006392848213052629
     },
     {
       "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": 9447.80204952202,
-      "required_pages_per_second_upper_estimate": 0.0037688694947829986
+      "required_token_positions_per_second": 9219.305407793161,
+      "required_pages_per_second_upper_estimate": 0.003677718768068119
     },
     {
       "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": 18895.60409904404,
-      "required_pages_per_second_upper_estimate": 0.004371165457039732
+      "required_token_positions_per_second": 18438.610815586322,
+      "required_pages_per_second_upper_estimate": 0.004265448103718881
     },
     {
       "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": 28343.406148566064,
-      "required_pages_per_second_upper_estimate": 0.0017259647280882448
+      "required_token_positions_per_second": 27657.91622337948,
+      "required_pages_per_second_upper_estimate": 0.0016842219881320682
     },
     {
       "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": 28343.406148566064,
-      "required_pages_per_second_upper_estimate": 0.0021819538648864705
+      "required_token_positions_per_second": 27657.91622337948,
+      "required_pages_per_second_upper_estimate": 0.0021291829528880457
     },
     {
       "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": 28343.406148566064,
-      "required_pages_per_second_upper_estimate": 0.35193024508692977
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27657.91622337948,
+      "required_pages_per_second_upper_estimate": 0.34341875440331127
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06372637284376381,
-    "producer_p95_seconds": 0.14824890799991408,
-    "gpu_update_seconds": 0.9754649760540074,
-    "producer_batches_per_second": 15.692090344631294,
-    "gpu_batches_per_second": 1.0251521321095942,
-    "producer_headroom": 15.307084532263085,
+    "producer_batch_seconds": 0.23146530392188325,
+    "producer_p95_seconds": 0.15556778400014082,
+    "gpu_update_seconds": 0.9996414688908812,
+    "producer_batches_per_second": 4.3203019331894685,
+    "gpu_batches_per_second": 1.0003586596997787,
+    "producer_headroom": 4.318752970545634,
     "sustainable": true,
-    "buffer_seconds": 60.47882851534846,
+    "buffer_seconds": 60.978129602343756,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 145,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 665687.3599506002,
+    "generation_bytes_per_second": 633170.2710066198,
     "upload_bytes_per_second": 4652435.752777713,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.42050551838162,
+    "estimated_full_shards_per_hour": 67.9318003542373,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 238800, "loss": 0.6271719429641962, "lr": 0.0003, "grad_norm": 0.14616646232827346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34953098704, "tokens_per_second": 149672.86953364583, "data_wait_s": 1.6234000213444233e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 238900, "loss": 0.6190658956766129, "lr": 0.0003, "grad_norm": 0.1668121387805161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34967725403, "tokens_per_second": 149104.00933536628, "data_wait_s": 1.2936999610246858e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239000, "loss": 0.6117204111069441, "lr": 0.0003, "grad_norm": 0.14575358029621122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34982211516, "tokens_per_second": 146874.45905564344, "data_wait_s": 1.5611999970133184e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 239100, "loss": 0.5797880366444588, "lr": 0.0003, "grad_norm": 0.15566681785785072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34996790354, "tokens_per_second": 147708.91423019825, "data_wait_s": 1.8267000086780172e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71eb4740b322488c60d3f6e85c1325ebab5e1d96e19743d63daad6ad2674b585
-size 40407
+oid sha256:8c62b7412978bffc51c4857e87554f641dc10bdbdc23b75bc1d1aa0ba2580de6
+size 42525