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": 75838.81615424603,
-      "required_pages_per_second_upper_estimate": 0.006573516372527788
+      "required_token_positions_per_second": 74833.39606963248,
+      "required_pages_per_second_upper_estimate": 0.006486369107807393
     },
     {
       "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": 9479.852019280754,
-      "required_pages_per_second_upper_estimate": 0.0037816547069095077
+      "required_token_positions_per_second": 9354.17450870406,
+      "required_pages_per_second_upper_estimate": 0.003731520068894232
     },
     {
       "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": 18959.704038561507,
-      "required_pages_per_second_upper_estimate": 0.004385993849926693
+      "required_token_positions_per_second": 18708.34901740812,
+      "required_pages_per_second_upper_estimate": 0.0043278472894801405
     },
     {
       "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": 28439.556057842263,
-      "required_pages_per_second_upper_estimate": 0.001731819753103578
+      "required_token_positions_per_second": 28062.523526112178,
+      "required_pages_per_second_upper_estimate": 0.0017088604500580314
     },
     {
       "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": 28439.556057842263,
-      "required_pages_per_second_upper_estimate": 0.0021893557510625364
+      "required_token_positions_per_second": 28062.523526112178,
+      "required_pages_per_second_upper_estimate": 0.0021603307430771075
     },
     {
       "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": 28439.556057842263,
-      "required_pages_per_second_upper_estimate": 0.3531241051670942
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28062.523526112178,
+      "required_pages_per_second_upper_estimate": 0.348442622969718
     }
   ],
   "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": 0.39527360431242187,
-    "producer_p95_seconds": 0.35628447199997026,
-    "gpu_update_seconds": 0.9721670740488233,
-    "producer_batches_per_second": 2.5298931906659927,
-    "gpu_batches_per_second": 1.0286297763976513,
-    "producer_headroom": 2.4594788608258,
+    "producer_batch_seconds": 0.10668362164071254,
+    "producer_p95_seconds": 0.3517403309997462,
+    "gpu_update_seconds": 0.9852285727002967,
+    "producer_batches_per_second": 9.373510053565528,
+    "gpu_batches_per_second": 1.0149928937395898,
+    "producer_headroom": 9.235049931266248,
     "sustainable": true,
-    "buffer_seconds": 60.274358591027045,
+    "buffer_seconds": 60.0989429347181,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360254.54597830126,
-    "upload_bytes_per_second": 2741143.6640597144,
+    "generation_bytes_per_second": 330078.32800613437,
+    "upload_bytes_per_second": 2808169.9437597515,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.30223927032259,
+    "estimated_full_shards_per_hour": 70.82712535989783,
     "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": 1.1246725650016742,
-      "bytes": 878351,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1035908939993533,
       "bytes": 6360663,
@@ -562,6 +556,12 @@
       "bytes": 11480592,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.264268914997956,
+      "bytes": 11538285,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -263,3 +263,4 @@
 {"step": 221300, "loss": 0.5787149202078581, "lr": 0.0003, "grad_norm": 0.1392222626005347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32391682260, "tokens_per_second": 150249.8014659225, "data_wait_s": 1.381599940941669e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221400, "loss": 0.5887596476823092, "lr": 0.0003, "grad_norm": 0.15217444645913925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32406219146, "tokens_per_second": 145521.58387926136, "data_wait_s": 1.332699685008265e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221500, "loss": 0.5614271368831396, "lr": 0.0003, "grad_norm": 0.15897538511002124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32420766381, "tokens_per_second": 135064.96236603995, "data_wait_s": 1.1476997315185145e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 221600, "loss": 0.558541651815176, "lr": 0.0003, "grad_norm": 0.14744477080345314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32435445354, "tokens_per_second": 148832.58113124082, "data_wait_s": 1.5354999050032347e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17e8b353c30cbb0d62a869ca6641ccfcd9ab657345f34151fd50ba2f4444a400
-size 565350
+oid sha256:dddc49cf0a55e6d8b487a9b30a36af09570285e5797192ee20bbddd1426b479f
+size 567468