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": 76017.49433858601,
-      "required_pages_per_second_upper_estimate": 0.005259066643231341
+      "required_token_positions_per_second": 76243.68538533368,
+      "required_pages_per_second_upper_estimate": 0.0052747150646808814
     },
     {
       "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": 9502.186792323251,
-      "required_pages_per_second_upper_estimate": 0.003978367181721496
+      "required_token_positions_per_second": 9530.46067316671,
+      "required_pages_per_second_upper_estimate": 0.003990204865204881
     },
     {
       "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": 19004.373584646502,
-      "required_pages_per_second_upper_estimate": 0.003958109838647609
+      "required_token_positions_per_second": 19060.92134633342,
+      "required_pages_per_second_upper_estimate": 0.003969887246142235
     },
     {
       "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": 28506.560376969755,
-      "required_pages_per_second_upper_estimate": 0.0017089247676019361
+      "required_token_positions_per_second": 28591.382019500128,
+      "required_pages_per_second_upper_estimate": 0.001714009695556483
     },
     {
       "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": 28506.560376969755,
-      "required_pages_per_second_upper_estimate": 0.002216259682201618
+      "required_token_positions_per_second": 28591.382019500128,
+      "required_pages_per_second_upper_estimate": 0.0022228541918173786
     },
     {
       "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": 28506.560376969755,
-      "required_pages_per_second_upper_estimate": 0.3656515485559415
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28591.382019500128,
+      "required_pages_per_second_upper_estimate": 0.3667395495119371
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.07293527498437857,
-    "producer_p95_seconds": 0.12601951999999983,
-    "gpu_update_seconds": 0.9698820073128367,
-    "producer_batches_per_second": 13.710786724450989,
-    "gpu_batches_per_second": 1.0310532543753528,
-    "producer_headroom": 13.297845350148718,
+    "producer_batch_seconds": 0.058196623499996214,
+    "producer_p95_seconds": 0.12805302399999619,
+    "gpu_update_seconds": 0.9670046722870299,
+    "producer_batches_per_second": 17.183127471305358,
+    "gpu_batches_per_second": 1.0341211667932626,
+    "producer_headroom": 16.6161645492559,
     "sustainable": true,
-    "buffer_seconds": 60.132684453395875,
+    "buffer_seconds": 60.92129435408288,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 92,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 668293.7642303993,
+    "generation_bytes_per_second": 839597.7279907386,
     "upload_bytes_per_second": 1946525.548627147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.70014236061088,
+    "estimated_full_shards_per_hour": 90.07906379600342,
     "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": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 557200, "loss": 0.5685006380081177, "lr": 0.0003, "grad_norm": 0.20223924821874198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81556112305, "tokens_per_second": 148940.87266347083, "data_wait_s": 1.2630000014723919e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 557300, "loss": 0.5449313521385193, "lr": 0.0003, "grad_norm": 0.16618840335759535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81570754119, "tokens_per_second": 152862.13565577025, "data_wait_s": 1.6536999964955612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 557400, "loss": 0.5944049954414368, "lr": 0.0003, "grad_norm": 0.15248265103769928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81585405695, "tokens_per_second": 151926.2680251599, "data_wait_s": 1.2938000054418808e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 557500, "loss": 0.6515524983406067, "lr": 0.0003, "grad_norm": 0.2547269670709619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81600042352, "tokens_per_second": 151738.7602942602, "data_wait_s": 1.4462999956776912e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7335250a2661ce2ca72fdc0b5e7994ea7b4713b7eb0f36d487b495df6ec1818d
-size 6519
+oid sha256:49c93fe94a31e5ebd944b9df9a6588b29ad047645d66c02727aaa12a0848ee55
+size 8637