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": 77487.75283785345,
-      "required_pages_per_second_upper_estimate": 0.0053607825376803895
+      "required_token_positions_per_second": 77068.39121065705,
+      "required_pages_per_second_upper_estimate": 0.005331770127257907
     },
     {
       "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": 9685.969104731681,
-      "required_pages_per_second_upper_estimate": 0.004055312998115811
+      "required_token_positions_per_second": 9633.548901332131,
+      "required_pages_per_second_upper_estimate": 0.004033365753610744
     },
     {
       "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": 19371.938209463362,
-      "required_pages_per_second_upper_estimate": 0.004034663856665956
+      "required_token_positions_per_second": 19267.097802664262,
+      "required_pages_per_second_upper_estimate": 0.004012828364757158
     },
     {
       "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": 29057.907314195043,
-      "required_pages_per_second_upper_estimate": 0.0017419771746305642
+      "required_token_positions_per_second": 28900.646703996394,
+      "required_pages_per_second_upper_estimate": 0.0017325496411722033
     },
     {
       "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": 29057.907314195043,
-      "required_pages_per_second_upper_estimate": 0.002259124481452002
+      "required_token_positions_per_second": 28900.646703996394,
+      "required_pages_per_second_upper_estimate": 0.0022468981607253785
     },
     {
       "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": 29057.907314195043,
-      "required_pages_per_second_upper_estimate": 0.372723635076449
+      "required_token_positions_per_second": 28900.646703996394,
+      "required_pages_per_second_upper_estimate": 0.3707064648221084
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 0.05700349705466934,
-    "producer_p95_seconds": 0.1404629449998538,
-    "gpu_update_seconds": 0.951479392547092,
-    "producer_batches_per_second": 17.542783367149347,
-    "gpu_batches_per_second": 1.0509949115377257,
-    "producer_headroom": 16.691596861760488,
+    "producer_batch_seconds": 0.05618902426559913,
+    "producer_p95_seconds": 0.14967549799985136,
+    "gpu_update_seconds": 0.9566567932951072,
+    "producer_batches_per_second": 17.79707003405351,
+    "gpu_batches_per_second": 1.0453069554396843,
+    "producer_headroom": 17.02568794882608,
     "sustainable": true,
-    "buffer_seconds": 60.89468112301389,
+    "buffer_seconds": 60.26937797759175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 353716.7327595308,
+    "upload_bytes_per_second": 34790.06957199078,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 75.89937674607698,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.4174159639997015,
+      "bytes": 49312,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 648100, "loss": 0.5829259157180786, "lr": 0.0001, "grad_norm": 0.21165426190517386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94865161079, "tokens_per_second": 153921.6830137373, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648200, "loss": 0.5496170520782471, "lr": 0.0001, "grad_norm": 0.17056241102373523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94879908023, "tokens_per_second": 154547.97673710226, "data_wait_s": 1.6008999864425277e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648300, "loss": 0.5251896977424622, "lr": 0.0001, "grad_norm": 0.17190561853882502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94894560854, "tokens_per_second": 152289.052323069, "data_wait_s": 1.3718999980483204e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 648400, "loss": 0.6239676475524902, "lr": 0.0001, "grad_norm": 0.21783978781609148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94909152120, "tokens_per_second": 152245.14727116865, "data_wait_s": 1.3509999917005189e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc117d4c44141cd6db64cec44a31ff93d652fb68fa317c07dd6bfec20c282a33
-size 70059
+oid sha256:44310fec6b38e6d87ff05a4a8072fed57c57b19b5e90550ee9ab82716f87a7e4
+size 72177