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": 75846.00584134708,
-      "required_pages_per_second_upper_estimate": 0.006574139556383612
+      "required_token_positions_per_second": 74817.5457905918,
+      "required_pages_per_second_upper_estimate": 0.006484995245792312
     },
     {
       "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": 9480.750730168385,
-      "required_pages_per_second_upper_estimate": 0.0037820132161195093
+      "required_token_positions_per_second": 9352.193223823975,
+      "required_pages_per_second_upper_estimate": 0.003730729704732717
     },
     {
       "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": 18961.50146033677,
-      "required_pages_per_second_upper_estimate": 0.004386409651820857
+      "required_token_positions_per_second": 18704.38644764795,
+      "required_pages_per_second_upper_estimate": 0.0043269306187049715
     },
     {
       "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": 28442.252190505154,
-      "required_pages_per_second_upper_estimate": 0.0017319839334372332
+      "required_token_positions_per_second": 28056.579671471925,
+      "required_pages_per_second_upper_estimate": 0.0017084985004954352
     },
     {
       "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": 28442.252190505154,
-      "required_pages_per_second_upper_estimate": 0.0021895633068182985
+      "required_token_positions_per_second": 28056.579671471925,
+      "required_pages_per_second_upper_estimate": 0.002159873168693258
     },
     {
       "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": 28442.252190505154,
-      "required_pages_per_second_upper_estimate": 0.3531575821113917
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28056.579671471925,
+      "required_pages_per_second_upper_estimate": 0.3483688201878879
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3665597756095167,
-    "producer_p95_seconds": 0.3851686769994558,
-    "gpu_update_seconds": 0.9720749192017115,
-    "producer_batches_per_second": 2.728068016566185,
-    "gpu_batches_per_second": 1.028727292770007,
-    "producer_headroom": 2.6518864967803477,
+    "producer_batch_seconds": 0.15711286308589933,
+    "producer_p95_seconds": 0.3503623950018664,
+    "gpu_update_seconds": 0.9854372957696134,
+    "producer_batches_per_second": 6.364851230884027,
+    "gpu_batches_per_second": 1.0147779105711778,
+    "producer_headroom": 6.272161784938251,
     "sustainable": true,
-    "buffer_seconds": 60.26864499050611,
+    "buffer_seconds": 60.11167504194641,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361905.6733767334,
-    "upload_bytes_per_second": 4258680.726722352,
+    "generation_bytes_per_second": 374399.8448116255,
+    "upload_bytes_per_second": 4221736.650489915,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.65653277374746,
+    "estimated_full_shards_per_hour": 80.33749111424993,
     "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": 1.2787357830020483,
-      "bytes": 10476048,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.3773436970004695,
       "bytes": 2708180,
@@ -562,6 +556,12 @@
       "bytes": 6167274,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.2202410780009814,
+      "bytes": 6552255,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 189700, "loss": 0.5790050812065601, "lr": 0.0003, "grad_norm": 0.14124361286865064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27764026242, "tokens_per_second": 152038.68240380072, "data_wait_s": 9.209004929289222e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 189800, "loss": 0.6157660949975252, "lr": 0.0003, "grad_norm": 0.16425549973474607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27778820913, "tokens_per_second": 151815.9570990028, "data_wait_s": 2.1535000996664166e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 189900, "loss": 0.6188155319541693, "lr": 0.0003, "grad_norm": 0.14061967457514632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27793615160, "tokens_per_second": 151529.41461533174, "data_wait_s": 1.0442003258503973e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 190000, "loss": 0.5794018357992172, "lr": 0.0003, "grad_norm": 0.143421867137207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27808256418, "tokens_per_second": 150552.32028538815, "data_wait_s": 7.763999747112393e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:269b31c23fe4acfb55ad598fc88ef1441ad2f785692d258830c9f5399fbe55d4
-size 744045
+oid sha256:aa773e5ef5024a29dfb9ce180e8243f1c9862baa8cf185976f8cdd49aac0a37a
+size 746163