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": 76629.54779606078,
-      "required_pages_per_second_upper_estimate": 0.005301409921579126
+      "required_token_positions_per_second": 77263.74762329577,
+      "required_pages_per_second_upper_estimate": 0.0053452853371735816
     },
     {
       "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": 9578.693474507598,
-      "required_pages_per_second_upper_estimate": 0.0040103989319108845
+      "required_token_positions_per_second": 9657.968452911971,
+      "required_pages_per_second_upper_estimate": 0.004043589709918999
     },
     {
       "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": 19157.386949015196,
-      "required_pages_per_second_upper_estimate": 0.003989978487211832
+      "required_token_positions_per_second": 19315.936905823943,
+      "required_pages_per_second_upper_estimate": 0.004023000261971556
     },
     {
       "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": 28736.08042352279,
-      "required_pages_per_second_upper_estimate": 0.0017226841439359716
+      "required_token_positions_per_second": 28973.905358735916,
+      "required_pages_per_second_upper_estimate": 0.0017369413856643486
     },
     {
       "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": 28736.08042352279,
-      "required_pages_per_second_upper_estimate": 0.002234103856269125
+      "required_token_positions_per_second": 28973.905358735916,
+      "required_pages_per_second_upper_estimate": 0.0022525936989006086
     },
     {
       "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": 28736.08042352279,
-      "required_pages_per_second_upper_estimate": 0.3685955852736983
+      "required_token_positions_per_second": 28973.905358735916,
+      "required_pages_per_second_upper_estimate": 0.371646148186092
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05974541284372492,
-    "producer_p95_seconds": 0.1932526880009391,
-    "gpu_update_seconds": 0.9621353919015305,
-    "producer_batches_per_second": 16.737686667521796,
-    "gpu_batches_per_second": 1.0393547606887583,
-    "producer_headroom": 16.103920721381105,
+    "producer_batch_seconds": 0.10749141742969925,
+    "producer_p95_seconds": 0.21705142599967076,
+    "gpu_update_seconds": 0.9542379481702786,
+    "producer_batches_per_second": 9.303068318491684,
+    "gpu_batches_per_second": 1.0479566463663164,
+    "producer_headroom": 8.87734082392543,
     "sustainable": true,
-    "buffer_seconds": 60.61452968979642,
+    "buffer_seconds": 60.11699073472755,
     "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": 309637.9098508861,
-    "upload_bytes_per_second": 5578208.903598402,
+    "generation_bytes_per_second": 336316.7066486268,
+    "upload_bytes_per_second": 5423696.2440244015,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.44108745236336,
+    "estimated_full_shards_per_hour": 72.16573619455436,
     "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": 1.0102191360001598,
-      "bytes": 6169276,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.2401527040001383,
       "bytes": 8401681,
@@ -562,6 +556,12 @@
       "bytes": 6774100,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 2.5239626219990896,
+      "bytes": 2432130,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -127,3 +127,4 @@
 {"step": 617100, "loss": 0.5785061717033386, "lr": 0.0001, "grad_norm": 0.14414073993283827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90325585542, "tokens_per_second": 153308.45878248534, "data_wait_s": 1.254800008609891e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 617200, "loss": 0.5823729038238525, "lr": 0.0001, "grad_norm": 0.14502301136742657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90340333233, "tokens_per_second": 153932.43073792415, "data_wait_s": 9.142999260802753e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 617300, "loss": 0.625805675983429, "lr": 0.0001, "grad_norm": 0.177189958780328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90355200921, "tokens_per_second": 154717.50190293862, "data_wait_s": 1.752900061546825e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 617400, "loss": 0.5926907062530518, "lr": 0.0001, "grad_norm": 0.15967770742117846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90370042945, "tokens_per_second": 154765.02048289645, "data_wait_s": 1.3943001249572262e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5ed6fe136ec7e245ff39215804f0bbc2cd9c4f0ca80dfd36d1890ed5654c67c4
-size 275736
+oid sha256:93cb3539c05ad0d5be9125df97a1679d129cee607e09ef888ac9e4be7a9ac7f9
+size 277854