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": 76084.10759480963,
-      "required_pages_per_second_upper_estimate": 0.00526367510286094
+      "required_token_positions_per_second": 76796.33059014821,
+      "required_pages_per_second_upper_estimate": 0.005312948342263494
     },
     {
       "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": 9510.513449351203,
-      "required_pages_per_second_upper_estimate": 0.003981853379138728
+      "required_token_positions_per_second": 9599.541323768526,
+      "required_pages_per_second_upper_estimate": 0.004019127491043837
     },
     {
       "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": 19021.026898702406,
-      "required_pages_per_second_upper_estimate": 0.0039615782847880286
+      "required_token_positions_per_second": 19199.082647537052,
+      "required_pages_per_second_upper_estimate": 0.0039986626016244205
     },
     {
       "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": 28531.540348053608,
-      "required_pages_per_second_upper_estimate": 0.0017104222787262058
+      "required_token_positions_per_second": 28798.623971305577,
+      "required_pages_per_second_upper_estimate": 0.0017264335341270816
     },
     {
       "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": 28531.540348053608,
-      "required_pages_per_second_upper_estimate": 0.002218201765078107
+      "required_token_positions_per_second": 28798.623971305577,
+      "required_pages_per_second_upper_estimate": 0.002238966342009243
     },
     {
       "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": 28531.540348053608,
-      "required_pages_per_second_upper_estimate": 0.3659719648036019
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28798.623971305577,
+      "required_pages_per_second_upper_estimate": 0.36939782675062627
     }
   ],
   "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": 0.0892404978517618,
-    "producer_p95_seconds": 0.3170936520000396,
-    "gpu_update_seconds": 0.9690328549641771,
-    "producer_batches_per_second": 11.205674823342077,
-    "gpu_batches_per_second": 1.0319567544868926,
-    "producer_headroom": 10.858667065863374,
+    "producer_batch_seconds": 0.11321709738291474,
+    "producer_p95_seconds": 0.3188606030016672,
+    "gpu_update_seconds": 0.9600458698147509,
+    "producer_batches_per_second": 8.832588214285973,
+    "gpu_batches_per_second": 1.0416168971103001,
+    "producer_headroom": 8.479689834899695,
     "sustainable": true,
-    "buffer_seconds": 60.080037007778984,
+    "buffer_seconds": 60.48288979832931,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317433.21362243436,
+    "generation_bytes_per_second": 318042.21452780213,
     "upload_bytes_per_second": 5719284.803494768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.1137781763532,
+    "estimated_full_shards_per_hour": 68.24445559382961,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 429400, "loss": 0.5377009510993958, "lr": 0.0003, "grad_norm": 0.14921407700600425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62852273036, "tokens_per_second": 152181.0550588608, "data_wait_s": 1.119099761126563e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429500, "loss": 0.597822368144989, "lr": 0.0003, "grad_norm": 0.16451327603277427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62866986073, "tokens_per_second": 152906.7777137519, "data_wait_s": 1.8337999790674075e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429600, "loss": 0.5384544730186462, "lr": 0.0003, "grad_norm": 0.14939019674607884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62881724585, "tokens_per_second": 152498.16433152763, "data_wait_s": 1.6655001672916114e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 429700, "loss": 0.5899538397789001, "lr": 0.0003, "grad_norm": 0.15180337830052834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62896493785, "tokens_per_second": 152997.20772784244, "data_wait_s": 1.1485000868560746e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0f7fd4fe7267edccc743afb59e24f68c86788db568ec51c3e6267573c9d916d
-size 693213
+oid sha256:fd65307bc2be72d0cb539fb7faa4ac703e4b30c92126c4319a1a92e37e31c105
+size 695331