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": 77502.90519226412,
-      "required_pages_per_second_upper_estimate": 0.005361830812716826
+      "required_token_positions_per_second": 77329.45701297477,
+      "required_pages_per_second_upper_estimate": 0.005349831265218365
     },
     {
       "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": 9687.863149033015,
-      "required_pages_per_second_upper_estimate": 0.004056105994912637
+      "required_token_positions_per_second": 9666.182126621847,
+      "required_pages_per_second_upper_estimate": 0.004047028603580312
     },
     {
       "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": 19375.72629806603,
-      "required_pages_per_second_upper_estimate": 0.004035452815623277
+      "required_token_positions_per_second": 19332.364253243693,
+      "required_pages_per_second_upper_estimate": 0.004026421645220806
     },
     {
       "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": 29063.58944709905,
-      "required_pages_per_second_upper_estimate": 0.0017423178098220447
+      "required_token_positions_per_second": 28998.54637986554,
+      "required_pages_per_second_upper_estimate": 0.001738418577256409
     },
     {
       "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": 29063.58944709905,
-      "required_pages_per_second_upper_estimate": 0.0022595662422922273
+      "required_token_positions_per_second": 28998.54637986554,
+      "required_pages_per_second_upper_estimate": 0.0022545094299090394
     },
     {
       "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": 29063.58944709905,
-      "required_pages_per_second_upper_estimate": 0.37279651937634267
+      "required_token_positions_per_second": 28998.54637986554,
+      "required_pages_per_second_upper_estimate": 0.37196221674767566
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0702952335859166,
-    "producer_p95_seconds": 0.1819795379997231,
-    "gpu_update_seconds": 0.951293371740071,
-    "producer_batches_per_second": 14.22571558536433,
-    "gpu_batches_per_second": 1.051200428497506,
-    "producer_headroom": 13.532828944616512,
+    "producer_batch_seconds": 0.0713370140704157,
+    "producer_p95_seconds": 0.1797094459998334,
+    "gpu_update_seconds": 0.9534271007182877,
+    "producer_batches_per_second": 14.017968273986277,
+    "gpu_batches_per_second": 1.0488478870032385,
+    "producer_headroom": 13.365110849427676,
     "sustainable": true,
-    "buffer_seconds": 60.882775791364544,
+    "buffer_seconds": 60.06590734525213,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297319.04967465816,
-    "upload_bytes_per_second": 5263450.416848346,
+    "generation_bytes_per_second": 284261.8861246741,
+    "upload_bytes_per_second": 5255606.408522934,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.79774682693299,
+    "estimated_full_shards_per_hour": 60.99598348431747,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 6182466,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.2363136280000617,
+      "bytes": 6022734,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 892100, "loss": 0.502311110496521, "lr": 3e-05, "grad_norm": 0.31728557763669496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130578377600, "tokens_per_second": 153429.09644075166, "data_wait_s": 2.1569000637100544e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 892200, "loss": 0.48435771465301514, "lr": 3e-05, "grad_norm": 0.304577815562576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130592973653, "tokens_per_second": 152639.74397078506, "data_wait_s": 1.3949000276625156e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 892300, "loss": 0.5495240092277527, "lr": 3e-05, "grad_norm": 0.330076856171934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130607484543, "tokens_per_second": 151557.41783641174, "data_wait_s": 1.48089993672329e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 892400, "loss": 0.5268421769142151, "lr": 3e-05, "grad_norm": 0.3321612467374445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130621985156, "tokens_per_second": 151539.6313387353, "data_wait_s": 1.5497000276809558e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8fc139dd0466958a71a528ae3d79c946df0c1ff4b592af8818e31c98d9783056
-size 155562
+oid sha256:c9e0b5b41335df0f0a06758288fa43af5bd67dc71442df642ab418be79a6202a
+size 157680