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": 77416.7751981749,
-      "required_pages_per_second_upper_estimate": 0.005355872139876613
+      "required_token_positions_per_second": 77349.60825024103,
+      "required_pages_per_second_upper_estimate": 0.005351225374569757
     },
     {
       "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": 9677.096899771863,
-      "required_pages_per_second_upper_estimate": 0.004051598391171842
+      "required_token_positions_per_second": 9668.701031280129,
+      "required_pages_per_second_upper_estimate": 0.004048083216360024
     },
     {
       "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": 19354.193799543726,
-      "required_pages_per_second_upper_estimate": 0.004030968164031254
+      "required_token_positions_per_second": 19337.402062560257,
+      "required_pages_per_second_upper_estimate": 0.004027470888045481
     },
     {
       "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": 29031.29069931559,
-      "required_pages_per_second_upper_estimate": 0.0017403815492097584
+      "required_token_positions_per_second": 29006.103093840386,
+      "required_pages_per_second_upper_estimate": 0.001738871590720767
     },
     {
       "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": 29031.29069931559,
-      "required_pages_per_second_upper_estimate": 0.0022570551567192404
+      "required_token_positions_per_second": 29006.103093840386,
+      "required_pages_per_second_upper_estimate": 0.0022550969311821113
     },
     {
       "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": 29031.29069931559,
-      "required_pages_per_second_upper_estimate": 0.37238222571946983
+      "required_token_positions_per_second": 29006.103093840386,
+      "required_pages_per_second_upper_estimate": 0.3720591461607776
     }
   ],
   "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": 0.10638115483595811,
-    "producer_p95_seconds": 0.22856816000057734,
-    "gpu_update_seconds": 0.952351732699635,
-    "producer_batches_per_second": 9.400161161457781,
-    "gpu_batches_per_second": 1.0500322156870512,
-    "producer_headroom": 8.95225976977013,
+    "producer_batch_seconds": 0.10950542241403127,
+    "producer_p95_seconds": 0.23196363699935318,
+    "gpu_update_seconds": 0.9531787124438379,
+    "producer_batches_per_second": 9.131967878440573,
+    "gpu_batches_per_second": 1.0491212056510555,
+    "producer_headroom": 8.704397384450472,
     "sustainable": true,
-    "buffer_seconds": 60.95051089277664,
+    "buffer_seconds": 60.05025888396179,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393433.2918256515,
+    "generation_bytes_per_second": 486830.31171840813,
     "upload_bytes_per_second": 4260532.603409401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.42162576748999,
+    "estimated_full_shards_per_hour": 104.46245206512626,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 699900, "loss": 0.5266990065574646, "lr": 8e-05, "grad_norm": 0.2204417780876645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102446083649, "tokens_per_second": 153643.0955539443, "data_wait_s": 1.6406000213464722e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700000, "loss": 0.5678468942642212, "lr": 8e-05, "grad_norm": 0.20413006877333337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102460742219, "tokens_per_second": 152822.87951846494, "data_wait_s": 1.8226999600301497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700100, "loss": 0.5273600816726685, "lr": 8e-05, "grad_norm": 0.20572507000430548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102475568774, "tokens_per_second": 145293.57576594577, "data_wait_s": 1.4528000974678434e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 700200, "loss": 0.5707261562347412, "lr": 8e-05, "grad_norm": 0.22813854833965583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102490078033, "tokens_per_second": 150402.06672966364, "data_wait_s": 1.755399898684118e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:604810906fd914bb5c766f87960ebeb74e068c0e056d52dc146d8b0c8608c554
-size 322332
+oid sha256:5f7c6ef802bc11f23cdb6e77e10a13d67c66b2eba4ef12094fc68024541b0bcb
+size 324450