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": 74551.71305228284,
-      "required_pages_per_second_upper_estimate": 0.006461953537782606
+      "required_token_positions_per_second": 75167.00011092928,
+      "required_pages_per_second_upper_estimate": 0.006515285060595287
     },
     {
       "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": 9318.964131535355,
-      "required_pages_per_second_upper_estimate": 0.003717474122999583
+      "required_token_positions_per_second": 9395.87501386616,
+      "required_pages_per_second_upper_estimate": 0.003748155023881506
     },
     {
       "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": 18637.92826307071,
-      "required_pages_per_second_upper_estimate": 0.004311556687327128
+      "required_token_positions_per_second": 18791.75002773232,
+      "required_pages_per_second_upper_estimate": 0.004347140645410994
     },
     {
       "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": 27956.89239460606,
-      "required_pages_per_second_upper_estimate": 0.0017024280683530246
+      "required_token_positions_per_second": 28187.625041598476,
+      "required_pages_per_second_upper_estimate": 0.001716478475994221
     },
     {
       "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": 27956.89239460606,
-      "required_pages_per_second_upper_estimate": 0.0021521989661680797
+      "required_token_positions_per_second": 28187.625041598476,
+      "required_pages_per_second_upper_estimate": 0.0021699614040424017
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27956.89239460606,
-      "required_pages_per_second_upper_estimate": 0.3471310378410676
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28187.625041598476,
+      "required_pages_per_second_upper_estimate": 0.3499959651042189
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.45146313383585834,
-    "producer_p95_seconds": 0.34129619099985575,
-    "gpu_update_seconds": 0.988951118377318,
-    "producer_batches_per_second": 2.21502028638196,
-    "gpu_batches_per_second": 1.0111723233002772,
-    "producer_headroom": 2.1905467894458863,
+    "producer_batch_seconds": 0.07839426777343306,
+    "producer_p95_seconds": 0.2590022440017492,
+    "gpu_update_seconds": 0.9808559592799282,
+    "producer_batches_per_second": 12.756034700012707,
+    "gpu_batches_per_second": 1.0195176881365189,
+    "producer_headroom": 12.511832652289016,
     "sustainable": true,
-    "buffer_seconds": 60.3260182210164,
+    "buffer_seconds": 60.81306947535555,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326764.8429471299,
-    "upload_bytes_per_second": 5478644.986612486,
+    "generation_bytes_per_second": 325122.7800168762,
+    "upload_bytes_per_second": 5430455.442252882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.11612860021995,
+    "estimated_full_shards_per_hour": 69.76378012065615,
     "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": 1.3106396379989746,
-      "bytes": 15458880,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.9607973950005544,
       "bytes": 7512102,
@@ -562,6 +556,12 @@
       "bytes": 2379532,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1314427079996676,
+      "bytes": 10762955,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -222,3 +222,4 @@
 {"step": 337200, "loss": 0.5765970721840858, "lr": 0.0003, "grad_norm": 0.1630461719034561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49357314524, "tokens_per_second": 146343.8726274883, "data_wait_s": 1.1293999705230817e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337300, "loss": 0.5907686352729797, "lr": 0.0003, "grad_norm": 0.14859957309270802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49371839460, "tokens_per_second": 146637.33944066992, "data_wait_s": 1.5142999473027885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337400, "loss": 0.5923497267067432, "lr": 0.0003, "grad_norm": 0.15710091453944813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49386590651, "tokens_per_second": 149696.4040228238, "data_wait_s": 1.3985998521093279e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 337500, "loss": 0.6300313007086515, "lr": 0.0003, "grad_norm": 0.1427950139887913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49401225614, "tokens_per_second": 148533.7911642823, "data_wait_s": 1.0522999218665063e-05, "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 sha256:689350759b32dbe406e4f37bec62ba526dfaa90539fe642955836236065705e2
-size 477729
+oid sha256:33316b5af38ec87acc1d0d3dfefcd63ab3acf1744d8a3f4a95403a9c818e4e96
+size 479847