Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75480.6289236848,
-      "required_pages_per_second_upper_estimate": 0.005221925048129088
+      "required_token_positions_per_second": 77352.40807721495,
+      "required_pages_per_second_upper_estimate": 0.005351419073096306
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9435.0786154606,
-      "required_pages_per_second_upper_estimate": 0.003950270389447193
+      "required_token_positions_per_second": 9669.051009651868,
+      "required_pages_per_second_upper_estimate": 0.004048229744995892
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18870.1572309212,
-      "required_pages_per_second_upper_estimate": 0.003930156111689898
+      "required_token_positions_per_second": 19338.102019303737,
+      "required_pages_per_second_upper_estimate": 0.0040276166705760464
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28305.2358463818,
-      "required_pages_per_second_upper_estimate": 0.001696855669397947
+      "required_token_positions_per_second": 29007.153028955607,
+      "required_pages_per_second_upper_estimate": 0.0017389345327277663
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28305.2358463818,
-      "required_pages_per_second_upper_estimate": 0.002200607585481429
+      "required_token_positions_per_second": 29007.153028955607,
+      "required_pages_per_second_upper_estimate": 0.0022551785590191435
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28305.2358463818,
-      "required_pages_per_second_upper_estimate": 0.3630691736429984
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29007.153028955607,
+      "required_pages_per_second_upper_estimate": 0.3720726136011032
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16358676684427564,
-    "producer_p95_seconds": 0.3577924040000653,
-    "gpu_update_seconds": 0.9767804144099435,
-    "producer_batches_per_second": 6.11296389855261,
-    "gpu_batches_per_second": 1.0237715511567491,
-    "producer_headroom": 5.971023410101242,
+    "producer_batch_seconds": 0.3733718101332215,
+    "producer_p95_seconds": 0.34329120199981844,
+    "gpu_update_seconds": 0.9531442114433337,
+    "producer_batches_per_second": 2.6782953957964675,
+    "gpu_batches_per_second": 1.0491591807347949,
+    "producer_headroom": 2.5528017530387355,
     "sustainable": true,
-    "buffer_seconds": 60.56038569341649,
+    "buffer_seconds": 60.04808532093003,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 492807.9753460714,
-    "upload_bytes_per_second": 2870095.3315929035,
+    "generation_bytes_per_second": 452432.6045715385,
+    "upload_bytes_per_second": 2856840.2837889856,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.74511952673537,
+    "estimated_full_shards_per_hour": 97.08150484904877,
     "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": 1.0191766480056685,
-      "bytes": 906488,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.433525214997644,
       "bytes": 7753087,
@@ -562,6 +556,12 @@
       "bytes": 6325909,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 2.196146342001157,
+      "bytes": 2310050,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -430,3 +430,4 @@
 {"step": 885500, "loss": 0.5140572190284729, "lr": 3e-05, "grad_norm": 0.3209864156546881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129612357157, "tokens_per_second": 151890.3066341143, "data_wait_s": 1.202000566991046e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 885600, "loss": 0.5332126021385193, "lr": 3e-05, "grad_norm": 0.307629663140919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129627117269, "tokens_per_second": 153366.5062806859, "data_wait_s": 1.2704003893304616e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 885700, "loss": 0.5637556910514832, "lr": 3e-05, "grad_norm": 0.3169692975692317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129641887771, "tokens_per_second": 153566.3688911708, "data_wait_s": 1.7531005141790956e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 885800, "loss": 0.5134627223014832, "lr": 3e-05, "grad_norm": 0.3256320482823105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129656413302, "tokens_per_second": 150971.0908989782, "data_wait_s": 1.4251003449317068e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7751fa9b4af792bf953af1ac9b11c1aa40d5270e8028ac60d7171e45abf86c34
-size 922188
+oid sha256:fe118a18cdcaf3fe864d5ae17b089a16a60e971fd5f9d42d66818ad60a470997
+size 924306