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": 75050.02360834069,
-      "required_pages_per_second_upper_estimate": 0.006505145833825144
+      "required_token_positions_per_second": 73650.35389860343,
+      "required_pages_per_second_upper_estimate": 0.006383826010815564
     },
     {
       "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": 9381.252951042587,
-      "required_pages_per_second_upper_estimate": 0.0037423220644018616
+      "required_token_positions_per_second": 9206.294237325428,
+      "required_pages_per_second_upper_estimate": 0.0036725284176342063
     },
     {
       "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": 18762.505902085173,
-      "required_pages_per_second_upper_estimate": 0.00434037553162156
+      "required_token_positions_per_second": 18412.588474650856,
+      "required_pages_per_second_upper_estimate": 0.0042594282931210205
     },
     {
       "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": 28143.758853127758,
-      "required_pages_per_second_upper_estimate": 0.0017138072552644586
+      "required_token_positions_per_second": 27618.882711976286,
+      "required_pages_per_second_upper_estimate": 0.0016818450520806206
     },
     {
       "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": 28143.758853127758,
-      "required_pages_per_second_upper_estimate": 0.0021665844634245444
+      "required_token_positions_per_second": 27618.882711976286,
+      "required_pages_per_second_upper_estimate": 0.0021261780451285526
     },
     {
       "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": 28143.758853127758,
-      "required_pages_per_second_upper_estimate": 0.34945129385410134
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27618.882711976286,
+      "required_pages_per_second_upper_estimate": 0.3429340888284427
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07463674224996453,
-    "producer_p95_seconds": 0.20810498600076244,
-    "gpu_update_seconds": 0.9823847675886171,
-    "producer_batches_per_second": 13.398226796273056,
-    "gpu_batches_per_second": 1.0179310927780585,
-    "producer_headroom": 13.162213917356288,
+    "producer_batch_seconds": 0.13599794917203667,
+    "producer_p95_seconds": 0.21661142199991446,
+    "gpu_update_seconds": 1.0010542529300466,
+    "producer_batches_per_second": 7.3530520576821745,
+    "gpu_batches_per_second": 0.9989468573486794,
+    "producer_headroom": 7.360804034358771,
     "sustainable": true,
-    "buffer_seconds": 60.907855590494265,
+    "buffer_seconds": 60.063255175802794,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353812.5308571568,
-    "upload_bytes_per_second": 6029943.16648851,
+    "generation_bytes_per_second": 362404.56383497675,
+    "upload_bytes_per_second": 6010014.759500051,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.91993278776195,
+    "estimated_full_shards_per_hour": 77.763583052511,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 6846709,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.0866875020001316,
+      "bytes": 5928612,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -81,3 +81,4 @@
 {"step": 123100, "loss": 0.53884132578969, "lr": 0.0003, "grad_norm": 0.13073770905537593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18014480667, "tokens_per_second": 148549.3480506966, "data_wait_s": 1.0952999218716286e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 123200, "loss": 0.6005491511896253, "lr": 0.0003, "grad_norm": 0.12964781555995203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18029136354, "tokens_per_second": 149765.2573419628, "data_wait_s": 1.1881998943863437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 123300, "loss": 0.6340554095804691, "lr": 0.0003, "grad_norm": 0.14613843072694627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18043625975, "tokens_per_second": 147550.44479308464, "data_wait_s": 8.282999260700308e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 123400, "loss": 0.6141817066818476, "lr": 0.0003, "grad_norm": 0.1323430557285076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18058221992, "tokens_per_second": 149106.1882848603, "data_wait_s": 1.0318999557057396e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:207407e9a78dddf70e11c055a7c755647392823742f84efe445125d464a7c5af
-size 176742
+oid sha256:8d5bf71b9fc83e33f22eb85b3e0fc6eb73bd270d7cc2c9ecd0bcc84f9973ea5d
+size 178860