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": 76530.00748262326,
-      "required_pages_per_second_upper_estimate": 0.005294523491730171
+      "required_token_positions_per_second": 76035.3723160381,
+      "required_pages_per_second_upper_estimate": 0.005260303483194099
     },
     {
       "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": 9566.250935327907,
-      "required_pages_per_second_upper_estimate": 0.004005189500586112
+      "required_token_positions_per_second": 9504.421539504763,
+      "required_pages_per_second_upper_estimate": 0.003979302823699567
     },
     {
       "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": 19132.501870655815,
-      "required_pages_per_second_upper_estimate": 0.0039847955816532206
+      "required_token_positions_per_second": 19008.843079009526,
+      "required_pages_per_second_upper_estimate": 0.003959040716454934
     },
     {
       "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": 28698.752805983724,
-      "required_pages_per_second_upper_estimate": 0.0017204464102604765
+      "required_token_positions_per_second": 28513.26461851429,
+      "required_pages_per_second_upper_estimate": 0.001709326676645747
     },
     {
       "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": 28698.752805983724,
-      "required_pages_per_second_upper_estimate": 0.0022312017981923
+      "required_token_positions_per_second": 28513.26461851429,
+      "required_pages_per_second_upper_estimate": 0.0022167809074928636
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28698.752805983724,
-      "required_pages_per_second_upper_estimate": 0.3681167866751802
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28513.26461851429,
+      "required_pages_per_second_upper_estimate": 0.36573754336802106
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10467412978931634,
-    "producer_p95_seconds": 0.30834341399895493,
-    "gpu_update_seconds": 0.9633868129013383,
-    "producer_batches_per_second": 9.553458930231928,
-    "gpu_batches_per_second": 1.03800465878124,
-    "producer_headroom": 9.203676350979965,
+    "producer_batch_seconds": 0.12977612657806503,
+    "producer_p95_seconds": 0.2811619759959285,
+    "gpu_update_seconds": 0.9696539617581196,
+    "producer_batches_per_second": 7.705577492316847,
+    "gpu_batches_per_second": 1.0312957399636244,
+    "producer_headroom": 7.471743743059227,
     "sustainable": true,
-    "buffer_seconds": 60.69336921278431,
+    "buffer_seconds": 60.118545629003414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 271518.3994944644,
-    "upload_bytes_per_second": 3779429.9869705383,
+    "generation_bytes_per_second": 311490.84409427,
+    "upload_bytes_per_second": 3800650.8113716273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.26152790666055,
+    "estimated_full_shards_per_hour": 66.83868400689197,
     "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": 2.1062088899998344,
-      "bytes": 6113756,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.8720440979996056,
       "bytes": 6184562,
@@ -562,6 +556,12 @@
       "bytes": 6923893,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.4826895270016394,
+      "bytes": 6163509,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 833200, "loss": 0.5130776166915894, "lr": 3e-05, "grad_norm": 0.29794373546723096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121956979522, "tokens_per_second": 150814.42181676612, "data_wait_s": 1.4227000065147877e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 833300, "loss": 0.7206534147262573, "lr": 3e-05, "grad_norm": 0.5096107104908736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121971679660, "tokens_per_second": 152731.32183292808, "data_wait_s": 9.300994861405343e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 833400, "loss": 0.5625774264335632, "lr": 3e-05, "grad_norm": 0.3185966074273557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121986334734, "tokens_per_second": 151613.43011473026, "data_wait_s": 7.2359980549663305e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 833500, "loss": 0.5352956056594849, "lr": 3e-05, "grad_norm": 0.2866640807895663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122001082959, "tokens_per_second": 152374.65815727136, "data_wait_s": 8.353999874088913e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2aef5df615e220e60bf47cbf3e3945de376fc861cfc9614672363fbc54270b3d
-size 712275
+oid sha256:8109a152b5fd3cfa131e304734e0f025bd96fe1b10f5da7903c80c2f7a13463e
+size 714393