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": 74569.67148891005,
-      "required_pages_per_second_upper_estimate": 0.006463510129554212
+      "required_token_positions_per_second": 76137.29607574556,
+      "required_pages_per_second_upper_estimate": 0.006599387855632922
     },
     {
       "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": 9321.208936113757,
-      "required_pages_per_second_upper_estimate": 0.003718369609108727
+      "required_token_positions_per_second": 9517.162009468195,
+      "required_pages_per_second_upper_estimate": 0.003796538219829342
     },
     {
       "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": 18642.417872227514,
-      "required_pages_per_second_upper_estimate": 0.0043125952793911316
+      "required_token_positions_per_second": 19034.32401893639,
+      "required_pages_per_second_upper_estimate": 0.004403255869119621
     },
     {
       "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": 27963.62680834127,
-      "required_pages_per_second_upper_estimate": 0.0017028381588167607
+      "required_token_positions_per_second": 28551.486028404586,
+      "required_pages_per_second_upper_estimate": 0.0017386357010596531
     },
     {
       "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": 27963.62680834127,
-      "required_pages_per_second_upper_estimate": 0.0021527173999794664
+      "required_token_positions_per_second": 28551.486028404586,
+      "required_pages_per_second_upper_estimate": 0.002197972430038413
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27963.62680834127,
-      "required_pages_per_second_upper_estimate": 0.34721465672102597
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28551.486028404586,
+      "required_pages_per_second_upper_estimate": 0.3545139007959644
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.32403751150013704,
-    "producer_p95_seconds": 0.3914885769991088,
-    "gpu_update_seconds": 0.9887129516316131,
-    "producer_batches_per_second": 3.0860624603938085,
-    "gpu_batches_per_second": 1.0114159001859544,
-    "producer_headroom": 3.0512299241354808,
+    "producer_batch_seconds": 0.1445629144688496,
+    "producer_p95_seconds": 0.36184723299811594,
+    "gpu_update_seconds": 0.9683559017731774,
+    "producer_batches_per_second": 6.917403427249523,
+    "gpu_batches_per_second": 1.0326781694301428,
+    "producer_headroom": 6.69850843372308,
     "sustainable": true,
-    "buffer_seconds": 60.3114900495284,
+    "buffer_seconds": 60.038065909937,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364097.5403122029,
-    "upload_bytes_per_second": 4798137.863036595,
+    "generation_bytes_per_second": 364831.44414586225,
+    "upload_bytes_per_second": 4825260.741959515,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.12685639404836,
+    "estimated_full_shards_per_hour": 78.28433507234479,
     "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": 1.0402059740008553,
-      "bytes": 6293747,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1847501439988264,
       "bytes": 2572744,
@@ -562,6 +556,12 @@
       "bytes": 6532415,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.337209769000765,
+      "bytes": 10101382,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -401,3 +401,4 @@
 {"step": 195100, "loss": 0.5759536698460579, "lr": 0.0003, "grad_norm": 0.1433815500195194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28555098636, "tokens_per_second": 135442.66935053776, "data_wait_s": 9.759001841302961e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 195200, "loss": 0.611761312931776, "lr": 0.0003, "grad_norm": 0.13731755191137826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28569686874, "tokens_per_second": 148886.4980394214, "data_wait_s": 1.1459997040219605e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 195300, "loss": 0.5979965794831514, "lr": 0.0003, "grad_norm": 0.13250483478838535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28584231055, "tokens_per_second": 149748.62037063212, "data_wait_s": 1.3190001482143998e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 195400, "loss": 0.5992383174598217, "lr": 0.0003, "grad_norm": 0.13717040894266003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28598726348, "tokens_per_second": 148196.41333551586, "data_wait_s": 1.0492993169464171e-05, "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 sha256:19fb419959154dc735c4c3bd9c500f6593ee82afb9323ce254b1b61cf3bfb40a
-size 859983
+oid sha256:8e4fbe352910cb533506a38e6aa91ac94cc3763f70546d1b752af2e98252e0c5
+size 862101