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": 75266.23897071424,
-      "required_pages_per_second_upper_estimate": 0.006523886833442857
+      "required_token_positions_per_second": 73490.3819642948,
+      "required_pages_per_second_upper_estimate": 0.0063699600489948525
     },
     {
       "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": 9408.27987133928,
-      "required_pages_per_second_upper_estimate": 0.003753103506996681
+      "required_token_positions_per_second": 9186.29774553685,
+      "required_pages_per_second_upper_estimate": 0.003664551518085547
     },
     {
       "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": 18816.55974267856,
-      "required_pages_per_second_upper_estimate": 0.004352879936327737
+      "required_token_positions_per_second": 18372.5954910737,
+      "required_pages_per_second_upper_estimate": 0.004250176620222916
     },
     {
       "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": 28224.839614017837,
-      "required_pages_per_second_upper_estimate": 0.0017187446482047884
+      "required_token_positions_per_second": 27558.893236610544,
+      "required_pages_per_second_upper_estimate": 0.0016781920077712997
     },
     {
       "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": 28224.839614017837,
-      "required_pages_per_second_upper_estimate": 0.002172826284843774
+      "required_token_positions_per_second": 27558.893236610544,
+      "required_pages_per_second_upper_estimate": 0.0021215598892534012
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28224.839614017837,
-      "required_pages_per_second_upper_estimate": 0.3504580455445055
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27558.893236610544,
+      "required_pages_per_second_upper_estimate": 0.3421892203162589
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09741216246845852,
-    "producer_p95_seconds": 0.28648294899903703,
-    "gpu_update_seconds": 0.9795626964791909,
-    "producer_batches_per_second": 10.265658565210417,
-    "gpu_batches_per_second": 1.0208637013171962,
-    "producer_headroom": 10.05585618527222,
+    "producer_batch_seconds": 0.451670192351628,
+    "producer_p95_seconds": 0.31580961999861756,
+    "gpu_update_seconds": 1.0032333215497595,
+    "producer_batches_per_second": 2.2140048578222182,
+    "gpu_batches_per_second": 0.996777099125092,
+    "producer_headroom": 2.221163447440287,
     "sustainable": true,
-    "buffer_seconds": 60.732887181709835,
+    "buffer_seconds": 60.19399929298557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337541.16306008596,
+    "generation_bytes_per_second": 310270.97963782534,
     "upload_bytes_per_second": 5912903.436221193,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.42847603656706,
+    "estimated_full_shards_per_hour": 66.57692949153014,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -165,3 +165,4 @@
 {"step": 101500, "loss": 0.6316697914153337, "lr": 0.0003, "grad_norm": 0.14569175651799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14853105249, "tokens_per_second": 148596.53276208186, "data_wait_s": 1.0170002497034147e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101600, "loss": 0.5911474358290434, "lr": 0.0003, "grad_norm": 0.14319383396650515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14867664916, "tokens_per_second": 147244.51369991162, "data_wait_s": 1.5435998648172244e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101700, "loss": 0.5779572483152151, "lr": 0.0003, "grad_norm": 0.14472449275624752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14882299425, "tokens_per_second": 147313.65997996804, "data_wait_s": 1.9014001736650243e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 101800, "loss": 0.6054339203983545, "lr": 0.0003, "grad_norm": 0.14104400304823012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14896947164, "tokens_per_second": 148862.45410511692, "data_wait_s": 1.686799805611372e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3cb6e16ad161539a9ebd2303f440965c673b809aadce01f407c1ffd61e7b8b44
-size 356220
+oid sha256:e454afdebfbfc1efa4af0761fa8c953e27b281b19e9eb4f79bdb0c49de781549
+size 358338