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": 76053.50941191216,
-      "required_pages_per_second_upper_estimate": 0.005261558249570528
+      "required_token_positions_per_second": 76290.06908604302,
+      "required_pages_per_second_upper_estimate": 0.0052779239967209665
     },
     {
       "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": 9506.68867648902,
-      "required_pages_per_second_upper_estimate": 0.003980252026611672
+      "required_token_positions_per_second": 9536.258635755377,
+      "required_pages_per_second_upper_estimate": 0.0039926323510655285
     },
     {
       "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": 19013.37735297804,
-      "required_pages_per_second_upper_estimate": 0.00395998508614573
+      "required_token_positions_per_second": 19072.517271510755,
+      "required_pages_per_second_upper_estimate": 0.0039723023715515626
     },
     {
       "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": 28520.066029467063,
-      "required_pages_per_second_upper_estimate": 0.0017097344108472125
+      "required_token_positions_per_second": 28608.775907266132,
+      "required_pages_per_second_upper_estimate": 0.001715052432569126
     },
     {
       "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": 28520.066029467063,
-      "required_pages_per_second_upper_estimate": 0.002217309687278197
+      "required_token_positions_per_second": 28608.775907266132,
+      "required_pages_per_second_upper_estimate": 0.0022242064900835586
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28520.066029467063,
-      "required_pages_per_second_upper_estimate": 0.3658247845649371
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28608.775907266132,
+      "required_pages_per_second_upper_estimate": 0.36696265962809776
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11900169568758656,
-    "producer_p95_seconds": 0.3142748890022631,
-    "gpu_update_seconds": 0.9694227205306594,
-    "producer_batches_per_second": 8.403241602752331,
-    "gpu_batches_per_second": 1.0315417400704232,
-    "producer_headroom": 8.146293335816583,
+    "producer_batch_seconds": 0.14257040306259228,
+    "producer_p95_seconds": 0.3073953570019512,
+    "gpu_update_seconds": 0.9664167418284363,
+    "producer_batches_per_second": 7.014078508012444,
+    "gpu_batches_per_second": 1.0347502859977622,
+    "producer_headroom": 6.778522898642246,
     "sustainable": true,
-    "buffer_seconds": 60.10420867290088,
+    "buffer_seconds": 60.88425473519148,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360970.4538571368,
-    "upload_bytes_per_second": 5091956.917859195,
+    "generation_bytes_per_second": 358066.11235102086,
+    "upload_bytes_per_second": 5109420.410021223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.45585643563821,
+    "estimated_full_shards_per_hour": 76.83265235803574,
     "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": 1.298020391001046,
-      "bytes": 6312632,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.6241299210014404,
       "bytes": 2390001,
@@ -562,6 +556,12 @@
       "bytes": 11979278,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.191631180998229,
+      "bytes": 7251240,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 953300, "loss": 0.4815470576286316, "lr": 3e-05, "grad_norm": 0.35038475854920687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139537409847, "tokens_per_second": 138880.933205589, "data_wait_s": 1.095400148187764e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 953400, "loss": 0.5547279119491577, "lr": 3e-05, "grad_norm": 0.3430259935499054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139551962961, "tokens_per_second": 151408.32961208528, "data_wait_s": 8.521998097421601e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 953500, "loss": 0.568813681602478, "lr": 3e-05, "grad_norm": 0.35661222057761044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139566538487, "tokens_per_second": 151904.02677224466, "data_wait_s": 1.3524000678444281e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 953600, "loss": 0.5567885041236877, "lr": 3e-05, "grad_norm": 0.33955340563023323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139581171073, "tokens_per_second": 151886.19815135363, "data_wait_s": 1.3534001482184976e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d9a9c182414b6c86e5fdf66605cc15551636d7f8f2253ce6e4933eba434bddb
-size 607710
+oid sha256:290218e9f0942b5ce53193dcc8bbe5032009a59e61ed75625b3f8312e4cbb946
+size 609828