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": 74985.94385695297,
-      "required_pages_per_second_upper_estimate": 0.006499591563383501
+      "required_token_positions_per_second": 73811.20511734065,
+      "required_pages_per_second_upper_estimate": 0.00639776818678202
     },
     {
       "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": 9373.242982119122,
-      "required_pages_per_second_upper_estimate": 0.003739126768038584
+      "required_token_positions_per_second": 9226.400639667581,
+      "required_pages_per_second_upper_estimate": 0.0036805491621459954
     },
     {
       "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": 18746.485964238243,
-      "required_pages_per_second_upper_estimate": 0.0043366696009419494
+      "required_token_positions_per_second": 18452.801279335163,
+      "required_pages_per_second_upper_estimate": 0.004268730817763544
     },
     {
       "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": 28119.728946357365,
-      "required_pages_per_second_upper_estimate": 0.0017123439600173187
+      "required_token_positions_per_second": 27679.201919002742,
+      "required_pages_per_second_upper_estimate": 0.0016855181753181107
     },
     {
       "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": 28119.728946357365,
-      "required_pages_per_second_upper_estimate": 0.00216473457468232
+      "required_token_positions_per_second": 27679.201919002742,
+      "required_pages_per_second_upper_estimate": 0.002130821584659701
     },
     {
       "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": 28119.728946357365,
-      "required_pages_per_second_upper_estimate": 0.34915292283493754
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27679.201919002742,
+      "required_pages_per_second_upper_estimate": 0.3436830515043116
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06403736174218011,
-    "producer_p95_seconds": 0.21111417900101515,
-    "gpu_update_seconds": 0.9832242712133797,
-    "producer_batches_per_second": 15.615883802741365,
-    "gpu_batches_per_second": 1.0170619555250784,
-    "producer_headroom": 15.3539159713032,
+    "producer_batch_seconds": 0.11793018941408207,
+    "producer_p95_seconds": 0.23719477799932065,
+    "gpu_update_seconds": 0.9988727305399178,
+    "producer_batches_per_second": 8.479592926699649,
+    "gpu_batches_per_second": 1.0011285416305968,
+    "producer_headroom": 8.47003414055945,
     "sustainable": true,
-    "buffer_seconds": 60.95990481522954,
+    "buffer_seconds": 60.931236562934984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289931.16962773673,
-    "upload_bytes_per_second": 4023636.305700489,
+    "generation_bytes_per_second": 290624.2487546611,
+    "upload_bytes_per_second": 4081283.6203331496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.21247974990918,
+    "estimated_full_shards_per_hour": 62.36119839649081,
     "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": [
-    {
-      "seconds": 1.5608764529997643,
-      "bytes": 2529116,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0529914669996288,
       "bytes": 5481895,
@@ -562,6 +556,12 @@
       "bytes": 9894024,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0132985079999344,
+      "bytes": 6275693,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 327500, "loss": 0.6007689274847507, "lr": 0.0003, "grad_norm": 0.13406897571219664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47936968395, "tokens_per_second": 137695.06942940524, "data_wait_s": 8.6880008893786e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 327600, "loss": 0.6296123526990414, "lr": 0.0003, "grad_norm": 0.15900981263965358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47951661711, "tokens_per_second": 149924.10939036196, "data_wait_s": 1.2664000678341836e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327700, "loss": 0.5703619085252285, "lr": 0.0003, "grad_norm": 0.15337562812187577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47966257208, "tokens_per_second": 148246.69603577672, "data_wait_s": 1.0547999409027398e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 327800, "loss": 0.5956343207508326, "lr": 0.0003, "grad_norm": 0.16043855212215086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47980866946, "tokens_per_second": 148502.69899712343, "data_wait_s": 9.469000360695645e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6674a0f0d7931685641d9feb1954c51a614a5e12dff4e420048b1c47cc627e1e
-size 270717
+oid sha256:db441ec9c6102bee56cd2f2a453fcf2f53031e3d6364b2bb2c1717fb874b5b82
+size 272835