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": 76262.73435534067,
-      "required_pages_per_second_upper_estimate": 0.005276032916625665
+      "required_token_positions_per_second": 76797.58282488566,
+      "required_pages_per_second_upper_estimate": 0.005313034974768215
     },
     {
       "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": 9532.841794417584,
-      "required_pages_per_second_upper_estimate": 0.003991201790949156
+      "required_token_positions_per_second": 9599.697853110707,
+      "required_pages_per_second_upper_estimate": 0.004019193026610703
     },
     {
       "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": 19065.683588835167,
-      "required_pages_per_second_upper_estimate": 0.0039708790956665685
+      "required_token_positions_per_second": 19199.395706221414,
+      "required_pages_per_second_upper_estimate": 0.00399872780349246
     },
     {
       "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": 28598.525383252752,
-      "required_pages_per_second_upper_estimate": 0.001714437929306168
+      "required_token_positions_per_second": 28799.093559332123,
+      "required_pages_per_second_upper_estimate": 0.0017264616852122538
     },
     {
       "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": 28598.525383252752,
-      "required_pages_per_second_upper_estimate": 0.0022234095569288086
+      "required_token_positions_per_second": 28799.093559332123,
+      "required_pages_per_second_upper_estimate": 0.002239002850412808
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28598.525383252752,
-      "required_pages_per_second_upper_estimate": 0.36683117691220934
+      "required_token_positions_per_second": 28799.093559332123,
+      "required_pages_per_second_upper_estimate": 0.36940385012162646
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4299784397812054,
+    "producer_batch_seconds": 0.25807861913278884,
     "producer_p95_seconds": 0.14699011099946802,
-    "gpu_update_seconds": 0.9667631330456857,
-    "producer_batches_per_second": 2.325698005948508,
-    "gpu_batches_per_second": 1.0343795349845468,
-    "producer_headroom": 2.248399090748883,
+    "gpu_update_seconds": 0.9600302156399253,
+    "producer_batches_per_second": 3.874788246156383,
+    "gpu_batches_per_second": 1.041633881630936,
+    "producer_headroom": 3.71991379551656,
     "sustainable": true,
-    "buffer_seconds": 60.9060773818782,
+    "buffer_seconds": 60.481903585315294,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 71,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319214.7702813808,
-    "upload_bytes_per_second": 5452637.346098469,
+    "generation_bytes_per_second": 327336.0598810324,
+    "upload_bytes_per_second": 5564567.110805273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.49605876284664,
+    "estimated_full_shards_per_hour": 70.23869845698574,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 6003921,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0415576699997473,
+      "bytes": 6858425,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 486300, "loss": 0.5967402458190918, "lr": 0.0003, "grad_norm": 0.15727533062311258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71178073082, "tokens_per_second": 152497.0445122778, "data_wait_s": 8.466000053886091e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 486400, "loss": 0.6144828796386719, "lr": 0.0003, "grad_norm": 0.15597788152546366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71192908927, "tokens_per_second": 153212.1731888354, "data_wait_s": 9.916999260894954e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 486500, "loss": 0.6309879422187805, "lr": 0.0003, "grad_norm": 0.15481581152410082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71207610359, "tokens_per_second": 153431.4123156743, "data_wait_s": 1.1533000360941514e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 486600, "loss": 0.5777984857559204, "lr": 0.0003, "grad_norm": 0.16132067041184758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71222190790, "tokens_per_second": 152174.30810914058, "data_wait_s": 1.1974999324593227e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1d0a7c45686c40ff37c0a352b5cb639eb412150070d7267b27b03d5ce9420dc
-size 31935
+oid sha256:d39049395bce529eb000440bad1795a33d32119e5c90dd2c27db240c751bfd80
+size 34053