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": 74340.04847959,
-      "required_pages_per_second_upper_estimate": 0.006443606989080546
+      "required_token_positions_per_second": 75407.0958612379,
+      "required_pages_per_second_upper_estimate": 0.006536095951714923
     },
     {
       "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": 9292.50605994875,
-      "required_pages_per_second_upper_estimate": 0.003706919602660264
+      "required_token_positions_per_second": 9425.886982654738,
+      "required_pages_per_second_upper_estimate": 0.0037601272469501906
     },
     {
       "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": 18585.0121198975,
-      "required_pages_per_second_upper_estimate": 0.004299315468896317
+      "required_token_positions_per_second": 18851.773965309476,
+      "required_pages_per_second_upper_estimate": 0.004361026126984248
     },
     {
       "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": 27877.51817984625,
-      "required_pages_per_second_upper_estimate": 0.0016975945951184727
+      "required_token_positions_per_second": 28277.660947964214,
+      "required_pages_per_second_upper_estimate": 0.0017219611903100014
     },
     {
       "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": 27877.51817984625,
-      "required_pages_per_second_upper_estimate": 0.00214608851939398
+      "required_token_positions_per_second": 28277.660947964214,
+      "required_pages_per_second_upper_estimate": 0.002176892617349777
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27877.51817984625,
-      "required_pages_per_second_upper_estimate": 0.34614547574215887
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28277.660947964214,
+      "required_pages_per_second_upper_estimate": 0.3511139097304868
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08986091703900456,
-    "producer_p95_seconds": 0.21898948200032464,
-    "gpu_update_seconds": 0.9917669077151862,
-    "producer_batches_per_second": 11.128308423181851,
-    "gpu_batches_per_second": 1.0083014387965223,
-    "producer_headroom": 11.036688032959924,
+    "producer_batch_seconds": 0.17497796114065522,
+    "producer_p95_seconds": 0.29304440899795736,
+    "gpu_update_seconds": 0.9777329196667681,
+    "producer_batches_per_second": 5.715005441148984,
+    "gpu_batches_per_second": 1.0227741951665297,
+    "producer_headroom": 5.587748955886061,
     "sustainable": true,
-    "buffer_seconds": 60.497781370626356,
+    "buffer_seconds": 60.619441019339625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352257.20914486673,
-    "upload_bytes_per_second": 5757753.5072629955,
+    "generation_bytes_per_second": 370439.1585049232,
+    "upload_bytes_per_second": 5918507.663151537,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.58619695434095,
+    "estimated_full_shards_per_hour": 79.48762003289006,
     "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": 0.9488432640009705,
-      "bytes": 921326,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.0550517599986051,
       "bytes": 5962735,
@@ -562,6 +556,12 @@
       "bytes": 1156879,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1430372309987433,
+      "bytes": 13489384,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -181,3 +181,4 @@
 {"step": 333100, "loss": 0.5879385620355606, "lr": 0.0003, "grad_norm": 0.15587732033841645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48757615375, "tokens_per_second": 150084.8878891177, "data_wait_s": 1.2298001820454374e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333200, "loss": 0.570095119997859, "lr": 0.0003, "grad_norm": 0.16794487178184855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48772284324, "tokens_per_second": 148634.77690898417, "data_wait_s": 1.9146998965879902e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333300, "loss": 0.5874333307147026, "lr": 0.0003, "grad_norm": 0.13781476407410762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48787016031, "tokens_per_second": 149138.41870287724, "data_wait_s": 1.671800055191852e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 333400, "loss": 0.672693058848381, "lr": 0.0003, "grad_norm": 0.17768767211908643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48801698217, "tokens_per_second": 149498.3728157396, "data_wait_s": 1.829400207498111e-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 sha256:980e8d1e5ac4b71693d0782469e03c176e4c1f8caa21482129b60b8e11a06da0
-size 390108
+oid sha256:2aa73aafd5126e840203dc7785d847422b404bb5fee03344b49c6579404a5ad5
+size 392226