Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 73993.46919830525,
-      "required_pages_per_second_upper_estimate": 0.006413566375375942
+      "required_token_positions_per_second": 73060.69997776051,
+      "required_pages_per_second_upper_estimate": 0.006332716303421099
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9249.183649788156,
-      "required_pages_per_second_upper_estimate": 0.0036896376455194495
+      "required_token_positions_per_second": 9132.587497220064,
+      "required_pages_per_second_upper_estimate": 0.0036431256969922067
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18498.36729957631,
-      "required_pages_per_second_upper_estimate": 0.004279271714611767
+      "required_token_positions_per_second": 18265.174994440127,
+      "required_pages_per_second_upper_estimate": 0.0042253267788629105
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27747.550949364468,
-      "required_pages_per_second_upper_estimate": 0.0016896802726674915
+      "required_token_positions_per_second": 27397.762491660193,
+      "required_pages_per_second_upper_estimate": 0.0016683799907915072
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27747.550949364468,
-      "required_pages_per_second_upper_estimate": 0.002136083282219172
+      "required_token_positions_per_second": 27397.762491660193,
+      "required_pages_per_second_upper_estimate": 0.0021091555984686736
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27747.550949364468,
-      "required_pages_per_second_upper_estimate": 0.3445317177119146
+      "required_token_positions_per_second": 27397.762491660193,
+      "required_pages_per_second_upper_estimate": 0.3401885157338887
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07634176457818853,
-    "producer_p95_seconds": 0.28655123399948934,
-    "gpu_update_seconds": 0.9964122617687545,
-    "producer_batches_per_second": 13.098989858635102,
-    "gpu_batches_per_second": 1.0036006564440272,
-    "producer_headroom": 13.051994111928579,
+    "producer_batch_seconds": 0.07679627486740515,
+    "producer_p95_seconds": 0.3275164560000121,
+    "gpu_update_seconds": 1.0091335016286815,
+    "producer_batches_per_second": 13.02146493077404,
+    "gpu_batches_per_second": 0.9909491641948854,
+    "producer_headroom": 13.140396501927086,
     "sustainable": true,
-    "buffer_seconds": 60.78114796789403,
+    "buffer_seconds": 60.54801009772089,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457420.60697074013,
+    "generation_bytes_per_second": 473659.0933884873,
     "upload_bytes_per_second": 2845999.660907743,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.15181404916433,
+    "estimated_full_shards_per_hour": 101.6362152217957,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 218400, "loss": 0.5609187074005604, "lr": 0.0003, "grad_norm": 0.15159725522328146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31966915344, "tokens_per_second": 149543.51853069908, "data_wait_s": 1.281100048799999e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218500, "loss": 0.5691029401496053, "lr": 0.0003, "grad_norm": 0.13410400851578652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31981575546, "tokens_per_second": 148790.43486262104, "data_wait_s": 1.061799775925465e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218600, "loss": 0.5861944854259491, "lr": 0.0003, "grad_norm": 0.14824412944482837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31996118978, "tokens_per_second": 147056.71789019948, "data_wait_s": 1.2038999557262287e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 218700, "loss": 0.5589394718408585, "lr": 0.0003, "grad_norm": 0.1404609947939712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32010632638, "tokens_per_second": 147477.66888674113, "data_wait_s": 1.5420999261550605e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:46d8150239045ab17d447ff5fa93e44c76a2e4158adfddfa2bc3789e52738dd4
-size 503145
+oid sha256:5b8001ef4eff6b95f5f191c6dfe58da2ef3e0ef38996116ae501e405ca16be8c
+size 505263