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": 75066.26136290142,
-      "required_pages_per_second_upper_estimate": 0.006506553281236255
+      "required_token_positions_per_second": 75834.5212433032,
+      "required_pages_per_second_upper_estimate": 0.006573144100532616
     },
     {
       "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": 9383.282670362678,
-      "required_pages_per_second_upper_estimate": 0.0037431317497856544
+      "required_token_positions_per_second": 9479.3151554129,
+      "required_pages_per_second_upper_estimate": 0.0037814405438857906
     },
     {
       "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": 18766.565340725356,
-      "required_pages_per_second_upper_estimate": 0.004341314611307286
+      "required_token_positions_per_second": 18958.6303108258,
+      "required_pages_per_second_upper_estimate": 0.004385745461911999
     },
     {
       "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": 28149.848011088037,
-      "required_pages_per_second_upper_estimate": 0.0017141780530368967
+      "required_token_positions_per_second": 28437.945466238696,
+      "required_pages_per_second_upper_estimate": 0.0017317216765250418
     },
     {
       "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": 28149.848011088037,
-      "required_pages_per_second_upper_estimate": 0.002167053223660195
+      "required_token_positions_per_second": 28437.945466238696,
+      "required_pages_per_second_upper_estimate": 0.002189231763262491
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28149.848011088037,
-      "required_pages_per_second_upper_estimate": 0.34952690081686727
+      "required_token_positions_per_second": 28437.945466238696,
+      "required_pages_per_second_upper_estimate": 0.3531041070096812
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10214540728131993,
-    "producer_p95_seconds": 0.19350108399976307,
-    "gpu_update_seconds": 0.9821722656942815,
-    "producer_batches_per_second": 9.78996537011094,
-    "gpu_batches_per_second": 1.0181513314195614,
-    "producer_headroom": 9.615432468630416,
+    "producer_batch_seconds": 0.07213013741412055,
+    "producer_p95_seconds": 0.1921677020000061,
+    "gpu_update_seconds": 0.9722221330237617,
+    "producer_batches_per_second": 13.863830513155726,
+    "gpu_batches_per_second": 1.0285715229397676,
+    "producer_headroom": 13.478722873380173,
     "sustainable": true,
-    "buffer_seconds": 60.894680473045455,
+    "buffer_seconds": 60.27777224747322,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 548862.6384725183,
-    "upload_bytes_per_second": 5382156.834192129,
+    "generation_bytes_per_second": 527975.9162604832,
+    "upload_bytes_per_second": 5390505.923136106,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.77314534789717,
+    "estimated_full_shards_per_hour": 113.29134097920296,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 2359249,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9851995050012192,
+      "bytes": 5930447,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 322500, "loss": 0.5936872847378254, "lr": 0.0003, "grad_norm": 0.1444422420849158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47204580366, "tokens_per_second": 149559.785035013, "data_wait_s": 1.221999991685152e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322600, "loss": 0.5737820789217949, "lr": 0.0003, "grad_norm": 0.1563632274313187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47219329241, "tokens_per_second": 149344.13109132333, "data_wait_s": 1.4562998330802657e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322700, "loss": 0.5852130390703678, "lr": 0.0003, "grad_norm": 0.16920465489237846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47233963042, "tokens_per_second": 149422.68517528864, "data_wait_s": 1.4547000318998471e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 322800, "loss": 0.5900401435792446, "lr": 0.0003, "grad_norm": 0.14580257372972075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47248730616, "tokens_per_second": 150173.88325679913, "data_wait_s": 1.47360005939845e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:150abb84d95c849deac9849bfe892693d661f8ed3e0e5215a0e8a1e3379f3278
-size 164034
+oid sha256:d12ade7f779c1a9b1b8e3f10bf355315da57bf5ec2747e5a2e4de0013172b254
+size 166152