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": 75808.40257977124,
-      "required_pages_per_second_upper_estimate": 0.006570880200974813
+      "required_token_positions_per_second": 74262.3352246316,
+      "required_pages_per_second_upper_estimate": 0.006436871001103197
     },
     {
       "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": 9476.050322471405,
-      "required_pages_per_second_upper_estimate": 0.0037801381532118258
+      "required_token_positions_per_second": 9282.79190307895,
+      "required_pages_per_second_upper_estimate": 0.003703044480245313
     },
     {
       "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": 18952.10064494281,
-      "required_pages_per_second_upper_estimate": 0.004384234938628164
+      "required_token_positions_per_second": 18565.5838061579,
+      "required_pages_per_second_upper_estimate": 0.004294821070439305
     },
     {
       "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": 28428.150967414214,
-      "required_pages_per_second_upper_estimate": 0.0017311252429343948
+      "required_token_positions_per_second": 27848.37570923685,
+      "required_pages_per_second_upper_estimate": 0.0016958199715570868
     },
     {
       "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": 28428.150967414214,
-      "required_pages_per_second_upper_estimate": 0.0021884777556300744
+      "required_token_positions_per_second": 27848.37570923685,
+      "required_pages_per_second_upper_estimate": 0.002143845051334947
     },
     {
       "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": 28428.150967414214,
-      "required_pages_per_second_upper_estimate": 0.35298249211436006
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27848.37570923685,
+      "required_pages_per_second_upper_estimate": 0.3457836237907651
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.07451443481252795,
-    "producer_p95_seconds": 0.19837748699956137,
-    "gpu_update_seconds": 0.9725570977757765,
-    "producer_batches_per_second": 13.42021854578802,
-    "gpu_batches_per_second": 1.0282172658931645,
-    "producer_headroom": 13.051928800408247,
+    "producer_batch_seconds": 0.054213417218761606,
+    "producer_p95_seconds": 0.19985040799929266,
+    "gpu_update_seconds": 0.9928047613502144,
+    "producer_batches_per_second": 18.44561828605651,
+    "gpu_batches_per_second": 1.0072473853167263,
+    "producer_headroom": 18.312897660445486,
     "sustainable": true,
-    "buffer_seconds": 60.29854006209814,
+    "buffer_seconds": 60.56109044236308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445026.0654993804,
-    "upload_bytes_per_second": 5749184.011970775,
+    "generation_bytes_per_second": 466941.0574348595,
+    "upload_bytes_per_second": 5653894.148586235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.49223397956905,
+    "estimated_full_shards_per_hour": 100.19468109402027,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 10590048,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9229088270003558,
+      "bytes": 1208028,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -69,3 +69,4 @@
 {"step": 361900, "loss": 0.6227873042225838, "lr": 0.0003, "grad_norm": 0.15211010385266396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52972141502, "tokens_per_second": 147651.26659480113, "data_wait_s": 1.0863999705179594e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 362000, "loss": 0.5736682824790478, "lr": 0.0003, "grad_norm": 0.15145500118993704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52986652671, "tokens_per_second": 146843.54463762057, "data_wait_s": 8.873999831848778e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 362100, "loss": 0.5817650128155947, "lr": 0.0003, "grad_norm": 0.1426328456273929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53001223874, "tokens_per_second": 147798.26814777747, "data_wait_s": 1.5081000128702726e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 362200, "loss": 0.5812425687909126, "lr": 0.0003, "grad_norm": 0.14682848256624062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53015925490, "tokens_per_second": 150572.19699902638, "data_wait_s": 1.411500034009805e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d60e63fa22adcb6d093acb8d9f54de001e3198279a817de4de4c3a51d8c913a4
-size 151326
+oid sha256:1be58cf86cfac5bf048907a000fcc5f5383d1f51994125b175d7bbd0a138b767
+size 153444