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": 75727.48836524444,
-      "required_pages_per_second_upper_estimate": 0.006563866761934832
+      "required_token_positions_per_second": 76415.77537678949,
+      "required_pages_per_second_upper_estimate": 0.0066235257356480745
     },
     {
       "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": 9465.936045655555,
-      "required_pages_per_second_upper_estimate": 0.003776103416968069
+      "required_token_positions_per_second": 9551.971922098686,
+      "required_pages_per_second_upper_estimate": 0.0038104244144322188
     },
     {
       "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": 18931.87209131111,
-      "required_pages_per_second_upper_estimate": 0.004379555418755855
+      "required_token_positions_per_second": 19103.943844197373,
+      "required_pages_per_second_upper_estimate": 0.004419361190426672
     },
     {
       "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": 28397.808136966665,
-      "required_pages_per_second_upper_estimate": 0.0017292775237566671
+      "required_token_positions_per_second": 28655.915766296057,
+      "required_pages_per_second_upper_estimate": 0.001744994924196753
     },
     {
       "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": 28397.808136966665,
-      "required_pages_per_second_upper_estimate": 0.0021861418805478915
+      "required_token_positions_per_second": 28655.915766296057,
+      "required_pages_per_second_upper_estimate": 0.002206011720341315
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28397.808136966665,
-      "required_pages_per_second_upper_estimate": 0.3526057357111224
+      "required_token_positions_per_second": 28655.915766296057,
+      "required_pages_per_second_upper_estimate": 0.35581056863672667
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12219626692132124,
-    "producer_p95_seconds": 0.35573608899721876,
-    "gpu_update_seconds": 0.9735962672418155,
-    "producer_batches_per_second": 8.183556054489554,
-    "gpu_batches_per_second": 1.0271197966206114,
-    "producer_headroom": 7.967479627415188,
+    "producer_batch_seconds": 0.29122143924956845,
+    "producer_p95_seconds": 0.40000440400035586,
+    "gpu_update_seconds": 0.9648269566913814,
+    "producer_batches_per_second": 3.4338131237069693,
+    "gpu_batches_per_second": 1.0364552866860552,
+    "producer_headroom": 3.313035465993121,
     "sustainable": true,
-    "buffer_seconds": 60.36296856899256,
+    "buffer_seconds": 60.78409827155703,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408348.9120163363,
-    "upload_bytes_per_second": 4937298.501192445,
+    "generation_bytes_per_second": 387466.9974561408,
+    "upload_bytes_per_second": 4801514.935471273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.62217064254347,
+    "estimated_full_shards_per_hour": 83.14139788401764,
     "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": 0.9012639180000406,
-      "bytes": 5714275,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.098721825001121,
       "bytes": 5875494,
@@ -562,6 +556,12 @@
       "bytes": 6225655,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 3.324964125000406,
+      "bytes": 5816918,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 193500, "loss": 0.6188144702464342, "lr": 0.0003, "grad_norm": 0.15031023066255766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28320553337, "tokens_per_second": 150369.8714985932, "data_wait_s": 8.048999006859958e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 193600, "loss": 0.5718002542853355, "lr": 0.0003, "grad_norm": 0.14237183209317347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28335333420, "tokens_per_second": 151310.95929628066, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 193700, "loss": 0.6380964834243059, "lr": 0.0003, "grad_norm": 0.16539279547047653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28350130021, "tokens_per_second": 152427.43554488552, "data_wait_s": 8.078000973910093e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 193800, "loss": 0.5223269741982222, "lr": 0.0003, "grad_norm": 0.21479016209627066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28364727251, "tokens_per_second": 137657.85232819806, "data_wait_s": 1.0907002433668822e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c761114c4db7df05a7accdf4bf0a78e0750a7cee10e11e7fbc112c0f6d5dfbdb
-size 825312
+oid sha256:5c12fe2b4f387b439c3b80303b95868ee5a337bb51de96807e6f8c24212e3078
+size 827430