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": 75036.81955730471,
-      "required_pages_per_second_upper_estimate": 0.005191221286383308
+      "required_token_positions_per_second": 76791.09529925407,
+      "required_pages_per_second_upper_estimate": 0.005312586152691889
     },
     {
       "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": 9379.602444663089,
-      "required_pages_per_second_upper_estimate": 0.003927043675208456
+      "required_token_positions_per_second": 9598.886912406759,
+      "required_pages_per_second_upper_estimate": 0.0040188535026723276
     },
     {
       "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": 18759.204889326178,
-      "required_pages_per_second_upper_estimate": 0.003907047664945669
+      "required_token_positions_per_second": 19197.773824813517,
+      "required_pages_per_second_upper_estimate": 0.003998390008367084
     },
     {
       "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": 28138.807333989265,
-      "required_pages_per_second_upper_estimate": 0.001686878534201639
+      "required_token_positions_per_second": 28796.660737220278,
+      "required_pages_per_second_upper_estimate": 0.0017263158412413528
     },
     {
       "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": 28138.807333989265,
-      "required_pages_per_second_upper_estimate": 0.0021876685006845704
+      "required_token_positions_per_second": 28796.660737220278,
+      "required_pages_per_second_upper_estimate": 0.0022388137091944654
     },
     {
       "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": 28138.807333989265,
-      "required_pages_per_second_upper_estimate": 0.36093440738304106
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28796.660737220278,
+      "required_pages_per_second_upper_estimate": 0.3693726444917366
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09231743015618576,
-    "producer_p95_seconds": 0.2541072090025409,
-    "gpu_update_seconds": 0.9825576355044582,
-    "producer_batches_per_second": 10.83219060916412,
-    "gpu_batches_per_second": 1.017752001374033,
-    "producer_headroom": 10.643251592273895,
+    "producer_batch_seconds": 0.12849819753111547,
+    "producer_p95_seconds": 0.2744159789981495,
+    "gpu_update_seconds": 0.9601113216666955,
+    "producer_batches_per_second": 7.782210328342177,
+    "gpu_batches_per_second": 1.0415458889330251,
+    "producer_headroom": 7.471788243832816,
     "sustainable": true,
-    "buffer_seconds": 60.918573401276404,
+    "buffer_seconds": 60.48701326500182,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276347.19114388374,
+    "generation_bytes_per_second": 359792.7056954616,
     "upload_bytes_per_second": 3945216.472421042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.29767418610939,
+    "estimated_full_shards_per_hour": 77.20313909671675,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 822000, "loss": 0.5329545736312866, "lr": 3e-05, "grad_norm": 0.30049129939144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120317816022, "tokens_per_second": 151462.89790822644, "data_wait_s": 1.4623001334257424e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822100, "loss": 0.5251942873001099, "lr": 3e-05, "grad_norm": 0.286140653402896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120332322134, "tokens_per_second": 148917.0294094196, "data_wait_s": 1.5284000255633146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822200, "loss": 0.508391261100769, "lr": 3e-05, "grad_norm": 0.29185700279355326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120346889303, "tokens_per_second": 151126.83623996298, "data_wait_s": 1.9220999092794955e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 822300, "loss": 0.5214368104934692, "lr": 3e-05, "grad_norm": 0.28538880079113754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120361651628, "tokens_per_second": 152933.67619670348, "data_wait_s": 1.2795997463399544e-05, "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 sha256:020773af261eaf4f8ce9ba4c49eed603551689d5635fcfaf9573ac13117b47cc
-size 473493
+oid sha256:e54f5d29f55b6b033f1b50934b5f209dcef37ef4f30ca1492db82784aad9b5a2
+size 475611