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": 74395.56145044182,
-      "required_pages_per_second_upper_estimate": 0.006448418712697644
+      "required_token_positions_per_second": 74931.84959177274,
+      "required_pages_per_second_upper_estimate": 0.0064949028095783404
     },
     {
       "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": 9299.445181305227,
-      "required_pages_per_second_upper_estimate": 0.0037096877219184726
+      "required_token_positions_per_second": 9366.481198971593,
+      "required_pages_per_second_upper_estimate": 0.003736429391643367
     },
     {
       "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": 18598.890362610455,
-      "required_pages_per_second_upper_estimate": 0.00430252595609923
+      "required_token_positions_per_second": 18732.962397943185,
+      "required_pages_per_second_upper_estimate": 0.0043335411618862256
     },
     {
       "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": 27898.33554391568,
-      "required_pages_per_second_upper_estimate": 0.0016988622633700326
+      "required_token_positions_per_second": 28099.443596914778,
+      "required_pages_per_second_upper_estimate": 0.0017111086886652685
     },
     {
       "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": 27898.33554391568,
-      "required_pages_per_second_upper_estimate": 0.0021476910977061966
+      "required_token_positions_per_second": 28099.443596914778,
+      "required_pages_per_second_upper_estimate": 0.002163172952328789
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27898.33554391568,
-      "required_pages_per_second_upper_estimate": 0.34640395773266547
+      "required_token_positions_per_second": 28099.443596914778,
+      "required_pages_per_second_upper_estimate": 0.34890104668555794
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.05207734765628658,
-    "producer_p95_seconds": 0.1905860759998177,
-    "gpu_update_seconds": 0.9910268645410183,
-    "producer_batches_per_second": 19.20220681360457,
-    "gpu_batches_per_second": 1.0090543816520428,
-    "producer_headroom": 19.029902810754713,
+    "producer_batch_seconds": 0.23628611993753168,
+    "producer_p95_seconds": 0.2043264320000162,
+    "gpu_update_seconds": 0.9839340734503247,
+    "producer_batches_per_second": 4.2321571841138015,
+    "gpu_batches_per_second": 1.016328255096744,
+    "producer_headroom": 4.164163657647149,
     "sustainable": true,
-    "buffer_seconds": 60.45263873700212,
+    "buffer_seconds": 60.0199784804698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465279.3152191602,
-    "upload_bytes_per_second": 5695084.725803528,
+    "generation_bytes_per_second": 443998.2543192281,
+    "upload_bytes_per_second": 5617608.209122828,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.83810989791017,
+    "estimated_full_shards_per_hour": 95.2716896265281,
     "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": [
@@ -412,6 +412,12 @@
       "bytes": 7343882,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.8120560110000952,
+      "bytes": 1154123,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -72,3 +72,4 @@
 {"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}
 {"step": 362300, "loss": 0.581356992945075, "lr": 0.0003, "grad_norm": 0.1601125700377772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53030612571, "tokens_per_second": 149150.71362877195, "data_wait_s": 7.561000529676676e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 362400, "loss": 0.6729161273688078, "lr": 0.0003, "grad_norm": 0.18195817486300464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53045348843, "tokens_per_second": 150079.46739753152, "data_wait_s": 8.346999493369367e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 362500, "loss": 0.558849785476923, "lr": 0.0003, "grad_norm": 0.13784731743430725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53060022456, "tokens_per_second": 149752.11108908063, "data_wait_s": 1.276300008612452e-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 sha256:2c4510e83bdbc96f9962826b6699a6c2fc11140bdbf2421d443bf74d6e9085f3
-size 157680
+oid sha256:d922ea7f74a6c6dac3aa3404965153a6ebf11efcdee46b18da2394676b9aca40
+size 159798