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": 76372.97621753161,
-      "required_pages_per_second_upper_estimate": 0.005283659704448397
+      "required_token_positions_per_second": 76120.17701833736,
+      "required_pages_per_second_upper_estimate": 0.005266170469273182
     },
     {
       "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": 9546.622027191452,
-      "required_pages_per_second_upper_estimate": 0.003996971286647591
+      "required_token_positions_per_second": 9515.02212729217,
+      "required_pages_per_second_upper_estimate": 0.0039837410684407095
     },
     {
       "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": 19093.244054382903,
-      "required_pages_per_second_upper_estimate": 0.003976619213821809
+      "required_token_positions_per_second": 19030.04425458434,
+      "required_pages_per_second_upper_estimate": 0.003963456362214569
     },
     {
       "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": 28639.866081574353,
-      "required_pages_per_second_upper_estimate": 0.0017169162410469565
+      "required_token_positions_per_second": 28545.06638187651,
+      "required_pages_per_second_upper_estimate": 0.001711233143800831
     },
     {
       "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": 28639.866081574353,
-      "required_pages_per_second_upper_estimate": 0.0022266236143847987
+      "required_token_positions_per_second": 28545.06638187651,
+      "required_pages_per_second_upper_estimate": 0.0022192533547131022
     },
     {
       "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": 28639.866081574353,
-      "required_pages_per_second_upper_estimate": 0.36736145100209533
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28545.06638187651,
+      "required_pages_per_second_upper_estimate": 0.3661454622423585
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06991982042193712,
-    "producer_p95_seconds": 0.15752997000072355,
-    "gpu_update_seconds": 0.9653676424760772,
-    "producer_batches_per_second": 14.302096229158124,
-    "gpu_batches_per_second": 1.0358747859365725,
-    "producer_headroom": 13.806780919208371,
+    "producer_batch_seconds": 0.08002289533593654,
+    "producer_p95_seconds": 0.16503802799979894,
+    "gpu_update_seconds": 0.9685736803034354,
+    "producer_batches_per_second": 12.496423627288099,
+    "gpu_batches_per_second": 1.032445977353751,
+    "producer_headroom": 12.10370702331324,
     "sustainable": true,
-    "buffer_seconds": 60.818161475992866,
+    "buffer_seconds": 60.05156817881299,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309662.2177731875,
-    "upload_bytes_per_second": 5930901.801681449,
+    "generation_bytes_per_second": 310633.71024878655,
+    "upload_bytes_per_second": 6035607.864857946,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.44630336662978,
+    "estimated_full_shards_per_hour": 66.65476303670594,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 5828242,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3037635500004399,
+      "bytes": 10399515,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 603800, "loss": 0.5839520692825317, "lr": 0.0003, "grad_norm": 0.16259424906493905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88378426504, "tokens_per_second": 151527.46729536954, "data_wait_s": 1.4981000276748091e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603900, "loss": 0.5592565536499023, "lr": 0.0003, "grad_norm": 0.17292819918029229, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88392975429, "tokens_per_second": 135907.64465507367, "data_wait_s": 1.5548999726888724e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 604000, "loss": 0.5574530363082886, "lr": 0.0003, "grad_norm": 0.17135670402991998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88407450940, "tokens_per_second": 150693.0444466796, "data_wait_s": 1.5667999832658097e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 604100, "loss": 0.557549774646759, "lr": 0.0003, "grad_norm": 0.1732923168330613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88422108370, "tokens_per_second": 152041.5890907132, "data_wait_s": 7.935999747132882e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fcba85e99c59f1b0bed708ba93d42d7a77b8a49cb3445016cc460f937f0fa13
-size 84885
+oid sha256:57676eb052eaf423a1762be5bc60063fedc563253fbab1acd884b21f24930050
+size 87003