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": 76643.98622186035,
-      "required_pages_per_second_upper_estimate": 0.005302408805377704
+      "required_token_positions_per_second": 75747.96771263136,
+      "required_pages_per_second_upper_estimate": 0.005240420165859864
     },
     {
       "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": 9580.498277732544,
-      "required_pages_per_second_upper_estimate": 0.004011154565332592
+      "required_token_positions_per_second": 9468.49596407892,
+      "required_pages_per_second_upper_estimate": 0.003964261535480091
     },
     {
       "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": 19160.996555465088,
-      "required_pages_per_second_upper_estimate": 0.003990730273043621
+      "required_token_positions_per_second": 18936.99192815784,
+      "required_pages_per_second_upper_estimate": 0.003944076016574798
     },
     {
       "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": 28741.494833197634,
-      "required_pages_per_second_upper_estimate": 0.0017230087295286535
+      "required_token_positions_per_second": 28405.48789223676,
+      "required_pages_per_second_upper_estimate": 0.0017028656264709413
     },
     {
       "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": 28741.494833197634,
-      "required_pages_per_second_upper_estimate": 0.0022345248028058724
+      "required_token_positions_per_second": 28405.48789223676,
+      "required_pages_per_second_upper_estimate": 0.0022084017410845046
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28741.494833197634,
-      "required_pages_per_second_upper_estimate": 0.368665035507467
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28405.48789223676,
+      "required_pages_per_second_upper_estimate": 0.36435509924496556
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.19990514079682953,
-    "producer_p95_seconds": 0.32467355900007533,
-    "gpu_update_seconds": 0.9619541419281157,
-    "producer_batches_per_second": 5.002372605396548,
-    "gpu_batches_per_second": 1.039550594372021,
-    "producer_headroom": 4.812053047228949,
+    "producer_batch_seconds": 0.11652121721093067,
+    "producer_p95_seconds": 0.32131955699878745,
+    "gpu_update_seconds": 0.9733330441247664,
+    "producer_batches_per_second": 8.582127992962569,
+    "gpu_batches_per_second": 1.0273975655467578,
+    "producer_headroom": 8.35326876445863,
     "sustainable": true,
-    "buffer_seconds": 60.603110941471286,
+    "buffer_seconds": 60.34664873573551,
     "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": 339792.6112575522,
+    "generation_bytes_per_second": 373502.47145661653,
     "upload_bytes_per_second": 5579985.104406778,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.91158440871168,
+    "estimated_full_shards_per_hour": 80.14493568204757,
     "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

@@ -329,3 +329,4 @@
 {"step": 430000, "loss": 0.5564310550689697, "lr": 0.0003, "grad_norm": 0.17130016792381678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62940228981, "tokens_per_second": 149703.11121873936, "data_wait_s": 1.0424002539366484e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430100, "loss": 0.566529393196106, "lr": 0.0003, "grad_norm": 0.15416684422087926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62954780166, "tokens_per_second": 138170.1454360266, "data_wait_s": 1.3216002116678283e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430200, "loss": 0.5586528182029724, "lr": 0.0003, "grad_norm": 0.1566217699200918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62969432541, "tokens_per_second": 151790.91207930402, "data_wait_s": 1.2141004845034331e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 430300, "loss": 0.5544184446334839, "lr": 0.0003, "grad_norm": 0.17044130918967768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62984072687, "tokens_per_second": 151063.73952058694, "data_wait_s": 1.5867997717577964e-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 sha256:1ad0eb43965746905c220aa6e23463b238c27120c274e317c78c6e6494f4bce5
-size 706704
+oid sha256:1b9c7f84f7a5439e5ad997512d1bc2ef177ba12239fac2e2837a485399f80e86
+size 708822