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": 74585.88375105569,
-      "required_pages_per_second_upper_estimate": 0.006464915367347393
+      "required_token_positions_per_second": 75138.92801960645,
+      "required_pages_per_second_upper_estimate": 0.006512851842867489
     },
     {
       "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": 9323.235468881961,
-      "required_pages_per_second_upper_estimate": 0.003719178023329329
+      "required_token_positions_per_second": 9392.366002450806,
+      "required_pages_per_second_upper_estimate": 0.0037467552267635255
     },
     {
       "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": 18646.470937763923,
-      "required_pages_per_second_upper_estimate": 0.004313532884771463
+      "required_token_positions_per_second": 18784.732004901613,
+      "required_pages_per_second_upper_estimate": 0.0043455171493420935
     },
     {
       "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": 27969.706406645888,
-      "required_pages_per_second_upper_estimate": 0.0017032083744563248
+      "required_token_positions_per_second": 28177.09800735242,
+      "required_pages_per_second_upper_estimate": 0.0017158374348397163
     },
     {
       "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": 27969.706406645888,
-      "required_pages_per_second_upper_estimate": 0.002153185424286361
+      "required_token_positions_per_second": 28177.09800735242,
+      "required_pages_per_second_upper_estimate": 0.0021691510038054445
     },
     {
       "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": 27969.706406645888,
-      "required_pages_per_second_upper_estimate": 0.3472901449848627
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28177.09800735242,
+      "required_pages_per_second_upper_estimate": 0.3498652545706001
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2597634689296342,
-    "producer_p95_seconds": 0.28857225899992045,
-    "gpu_update_seconds": 0.9884980413462815,
-    "producer_batches_per_second": 3.8496560125276282,
-    "gpu_batches_per_second": 1.011635793064449,
-    "producer_headroom": 3.8053774282404964,
+    "producer_batch_seconds": 0.07101085622663561,
+    "producer_p95_seconds": 0.27071953899940127,
+    "gpu_update_seconds": 0.9812224095180292,
+    "producer_batches_per_second": 14.082353785573817,
+    "gpu_batches_per_second": 1.0191369360298184,
+    "producer_headroom": 13.81792111316608,
     "sustainable": true,
-    "buffer_seconds": 60.298380522123175,
+    "buffer_seconds": 60.83578939011781,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385057.6391975924,
-    "upload_bytes_per_second": 6217736.415972115,
+    "generation_bytes_per_second": 384102.8301693622,
+    "upload_bytes_per_second": 6454944.819267293,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.62440568872289,
+    "estimated_full_shards_per_hour": 82.41952589808129,
     "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": 1.1883426349995716,
-      "bytes": 1212298,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1447565099988424,
       "bytes": 5933026,
@@ -562,6 +556,12 @@
       "bytes": 2069289,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.112150704000669,
+      "bytes": 16576455,
+      "pages": 44,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -155,3 +155,4 @@
 {"step": 55500, "loss": 0.6675822697579861, "lr": 0.0003, "grad_norm": 0.13384163295611565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8123233060, "tokens_per_second": 150093.26572923284, "data_wait_s": 7.4460003816057e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 55600, "loss": 0.643940506502986, "lr": 0.0003, "grad_norm": 0.15360191239520263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8137899645, "tokens_per_second": 149318.89784144715, "data_wait_s": 1.585300014994573e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 55700, "loss": 0.6222561784088612, "lr": 0.0003, "grad_norm": 0.15287018424892007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8152534973, "tokens_per_second": 149432.7195820138, "data_wait_s": 1.0710999049479142e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 55800, "loss": 0.6200186684727669, "lr": 0.0003, "grad_norm": 0.13037454080661143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8167305941, "tokens_per_second": 150812.42841196473, "data_wait_s": 1.304299985349644e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ab999dc55263fd3d789b6f3503c2e4350b76b35f96e619407d2ae7acc65ed7c
-size 335040
+oid sha256:eb840901012ef8878c6c293baadaf1646ae70e6c4af3b212bbf5f411639d4e02
+size 337158