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": 77475.49327565309,
-      "required_pages_per_second_upper_estimate": 0.0053599343927212655
+      "required_token_positions_per_second": 77123.19485064084,
+      "required_pages_per_second_upper_estimate": 0.005335561570233949
     },
     {
       "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": 9684.436659456636,
-      "required_pages_per_second_upper_estimate": 0.004054671395280245
+      "required_token_positions_per_second": 9640.399356330105,
+      "required_pages_per_second_upper_estimate": 0.004036233896064627
     },
     {
       "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": 19368.873318913273,
-      "required_pages_per_second_upper_estimate": 0.004034025520790972
+      "required_token_positions_per_second": 19280.79871266021,
+      "required_pages_per_second_upper_estimate": 0.004015681902992019
     },
     {
       "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": 29053.30997836991,
-      "required_pages_per_second_upper_estimate": 0.0017417015713675728
+      "required_token_positions_per_second": 28921.19806899031,
+      "required_pages_per_second_upper_estimate": 0.0017337816641234713
     },
     {
       "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": 29053.30997836991,
-      "required_pages_per_second_upper_estimate": 0.002258767058813141
+      "required_token_positions_per_second": 28921.19806899031,
+      "required_pages_per_second_upper_estimate": 0.002248495938957754
     },
     {
       "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": 29053.30997836991,
-      "required_pages_per_second_upper_estimate": 0.37266466538871884
+      "required_token_positions_per_second": 28921.19806899031,
+      "required_pages_per_second_upper_estimate": 0.37097007566591383
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.41775037589844644,
-    "producer_p95_seconds": 0.14299106599992228,
-    "gpu_update_seconds": 0.9516299526829762,
-    "producer_batches_per_second": 2.393774027969029,
-    "gpu_batches_per_second": 1.0508286305834025,
-    "producer_headroom": 2.277987064969904,
+    "producer_batch_seconds": 0.06019440963279976,
+    "producer_p95_seconds": 0.14486277699961647,
+    "gpu_update_seconds": 0.9559769942464641,
+    "producer_batches_per_second": 16.612838403104842,
+    "gpu_batches_per_second": 1.0460502773795686,
+    "producer_headroom": 15.881491322502397,
     "sustainable": true,
-    "buffer_seconds": 60.90431697171048,
+    "buffer_seconds": 60.22655063752724,
     "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": 306962.3323305865,
+    "generation_bytes_per_second": 292341.7376165368,
     "upload_bytes_per_second": 5099910.381886476,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.86697080076405,
+    "estimated_full_shards_per_hour": 62.729731525154854,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -21,3 +21,4 @@
 {"step": 399200, "loss": 0.5344240069389343, "lr": 0.0003, "grad_norm": 0.16703074589599587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58431709220, "tokens_per_second": 150203.48550130386, "data_wait_s": 8.62199976836564e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 399300, "loss": 0.5939231514930725, "lr": 0.0003, "grad_norm": 0.13250949760920044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58446294719, "tokens_per_second": 151739.47154631745, "data_wait_s": 1.1953000011999393e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 399400, "loss": 0.572610080242157, "lr": 0.0003, "grad_norm": 0.15060752239368838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58461052475, "tokens_per_second": 151394.76089819, "data_wait_s": 1.708799982225173e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 399500, "loss": 0.5737971067428589, "lr": 0.0003, "grad_norm": 0.14036649233757922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58475688757, "tokens_per_second": 152696.5091392368, "data_wait_s": 9.268999747291673e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3598a6e4f0d99be0869b50d2e7da4c1544d4ea5f3dd4534b078140c1e400db9
-size 48879
+oid sha256:f7e6a8c5a601e6cd49aeab75a6f42bca2975b6317e6ff12a768a072eda8d372a
+size 50997