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": 76126.84762455872,
-      "required_pages_per_second_upper_estimate": 0.005266631957289514
+      "required_token_positions_per_second": 76339.13990225257,
+      "required_pages_per_second_upper_estimate": 0.005281318829646324
     },
     {
       "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": 9515.85595306984,
-      "required_pages_per_second_upper_estimate": 0.003984090173881558
+      "required_token_positions_per_second": 9542.392487781572,
+      "required_pages_per_second_upper_estimate": 0.0039952004668195006
     },
     {
       "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": 19031.71190613968,
-      "required_pages_per_second_upper_estimate": 0.003963803690054618
+      "required_token_positions_per_second": 19084.784975563143,
+      "required_pages_per_second_upper_estimate": 0.0039748574107845635
     },
     {
       "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": 28547.567859209517,
-      "required_pages_per_second_upper_estimate": 0.00171138310354741
+      "required_token_positions_per_second": 28627.177463344717,
+      "required_pages_per_second_upper_estimate": 0.001716155577758488
     },
     {
       "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": 28547.567859209517,
-      "required_pages_per_second_upper_estimate": 0.0022194478335729076
+      "required_token_positions_per_second": 28627.177463344717,
+      "required_pages_per_second_upper_estimate": 0.0022256371301287775
     },
     {
       "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": 28547.567859209517,
-      "required_pages_per_second_upper_estimate": 0.3661775485077092
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28627.177463344717,
+      "required_pages_per_second_upper_estimate": 0.36719869503142233
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.26499354894554017,
-    "producer_p95_seconds": 0.31279561600240413,
-    "gpu_update_seconds": 0.9684888091466848,
-    "producer_batches_per_second": 3.773676770544757,
-    "gpu_batches_per_second": 1.032536453241085,
-    "producer_headroom": 3.654763721609399,
+    "producer_batch_seconds": 0.08533144863284292,
+    "producer_p95_seconds": 0.2975644199977978,
+    "gpu_update_seconds": 0.9657955289305594,
+    "producer_batches_per_second": 11.719008829941668,
+    "gpu_batches_per_second": 1.035415851538799,
+    "producer_headroom": 11.318166331455409,
     "sustainable": true,
-    "buffer_seconds": 60.04630616709446,
+    "buffer_seconds": 60.84511832262525,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456828.9360956987,
+    "generation_bytes_per_second": 456757.5227396627,
     "upload_bytes_per_second": 5153907.235389738,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.02485525277348,
+    "estimated_full_shards_per_hour": 98.009531608986,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 589400, "loss": 0.5678746700286865, "lr": 0.0003, "grad_norm": 0.1761332567785017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86270473152, "tokens_per_second": 150825.5584035463, "data_wait_s": 1.1059000826207921e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589500, "loss": 0.600641667842865, "lr": 0.0003, "grad_norm": 0.1509352730321945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86285036075, "tokens_per_second": 151553.0314019361, "data_wait_s": 1.591299951542169e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589600, "loss": 0.5750231146812439, "lr": 0.0003, "grad_norm": 0.20048132500355573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86299595829, "tokens_per_second": 150362.5483587851, "data_wait_s": 1.089699799194932e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 589700, "loss": 0.5951194763183594, "lr": 0.0003, "grad_norm": 0.153565967132763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86314303566, "tokens_per_second": 152970.13031007777, "data_wait_s": 1.2247000995557755e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa5df28ffb738340a27aa770a3eae65fb81f6c0268eb63a1423a1b70e29f769c
-size 693213
+oid sha256:dbcb56ca6e19d4ec62cfd7bd9ddcb0173e723c081620f04d67cdbbc97a0e7987
+size 695331