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": 74937.67691667979,
-      "required_pages_per_second_upper_estimate": 0.0064954079073320636
+      "required_token_positions_per_second": 75009.12139817621,
+      "required_pages_per_second_upper_estimate": 0.0065016005339137905
     },
     {
       "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": 9367.209614584974,
-      "required_pages_per_second_upper_estimate": 0.0037367199675223286
+      "required_token_positions_per_second": 9376.140174772027,
+      "required_pages_per_second_upper_estimate": 0.0037402825015047177
     },
     {
       "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": 18734.419229169947,
-      "required_pages_per_second_upper_estimate": 0.004333878174151183
+      "required_token_positions_per_second": 18752.280349544053,
+      "required_pages_per_second_upper_estimate": 0.00433801003000475
     },
     {
       "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": 28101.62884375492,
-      "required_pages_per_second_upper_estimate": 0.001711241758732729
+      "required_token_positions_per_second": 28128.420524316076,
+      "required_pages_per_second_upper_estimate": 0.001712873231513819
     },
     {
       "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": 28101.62884375492,
-      "required_pages_per_second_upper_estimate": 0.002163341178679693
+      "required_token_positions_per_second": 28128.420524316076,
+      "required_pages_per_second_upper_estimate": 0.0021654036764134094
     },
     {
       "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": 28101.62884375492,
-      "required_pages_per_second_upper_estimate": 0.3489281801377618
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28128.420524316076,
+      "required_pages_per_second_upper_estimate": 0.3492608431443445
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.1854412814610953,
-    "producer_p95_seconds": 0.29290093700183206,
-    "gpu_update_seconds": 0.9838575604895682,
-    "producer_batches_per_second": 5.392542545656402,
-    "gpu_batches_per_second": 1.0164072932492376,
-    "producer_headroom": 5.305493753805713,
+    "producer_batch_seconds": 0.09016516255479701,
+    "producer_p95_seconds": 0.28266207200067583,
+    "gpu_update_seconds": 0.9829204585482939,
+    "producer_batches_per_second": 11.09075802300317,
+    "gpu_batches_per_second": 1.017376321047312,
+    "producer_headroom": 10.901332961618447,
     "sustainable": true,
-    "buffer_seconds": 60.01531118986366,
+    "buffer_seconds": 60.941068429994225,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322165.4456866767,
-    "upload_bytes_per_second": 4853807.5599855585,
+    "generation_bytes_per_second": 321597.7275483063,
+    "upload_bytes_per_second": 4939741.117864166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.12920501661516,
+    "estimated_full_shards_per_hour": 69.00738591992274,
     "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": 1.0997196959979192,
-      "bytes": 878203,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.3262504000012996,
       "bytes": 7612878,
@@ -562,6 +556,12 @@
       "bytes": 2542185,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5899441080000543,
+      "bytes": 10502441,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -253,3 +253,4 @@
 {"step": 380300, "loss": 0.6201922632753849, "lr": 0.0003, "grad_norm": 0.15635784513159612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55664393099, "tokens_per_second": 147345.95866581547, "data_wait_s": 1.7039001249941066e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380400, "loss": 0.6774350851774216, "lr": 0.0003, "grad_norm": 0.2429773700394193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55678942383, "tokens_per_second": 147843.0419820773, "data_wait_s": 1.2770000466844067e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380500, "loss": 0.5221319552510977, "lr": 0.0003, "grad_norm": 0.1407149522118551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55693639735, "tokens_per_second": 150086.71387165552, "data_wait_s": 1.2709999282378703e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 380600, "loss": 0.5683339741080999, "lr": 0.0003, "grad_norm": 0.13495020868309707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55708230802, "tokens_per_second": 147894.6098215744, "data_wait_s": 8.877999789547175e-06, "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:cb7e2064413918fe3c74aba95f4c2de0aec422847ba9df5c9520fcef1dec2694
-size 544170
+oid sha256:e2d4aacd97687224718449d89070199d1d6edd92380d0b4e6841efc5970a4382
+size 546288