Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76530.53102406212,
-      "required_pages_per_second_upper_estimate": 0.005294559711541694
+      "required_token_positions_per_second": 75510.42381458056,
+      "required_pages_per_second_upper_estimate": 0.005223986327814929
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9566.316378007765,
-      "required_pages_per_second_upper_estimate": 0.00400521690006957
+      "required_token_positions_per_second": 9438.80297682257,
+      "required_pages_per_second_upper_estimate": 0.003951829701775941
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19132.63275601553,
-      "required_pages_per_second_upper_estimate": 0.003984822841621971
+      "required_token_positions_per_second": 18877.60595364514,
+      "required_pages_per_second_upper_estimate": 0.003931707484197268
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28698.949134023293,
-      "required_pages_per_second_upper_estimate": 0.0017204581798266729
+      "required_token_positions_per_second": 28316.408930467707,
+      "required_pages_per_second_upper_estimate": 0.0016975254787285857
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28698.949134023293,
-      "required_pages_per_second_upper_estimate": 0.002231217061833821
+      "required_token_positions_per_second": 28316.408930467707,
+      "required_pages_per_second_upper_estimate": 0.002201476243623905
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28698.949134023293,
-      "required_pages_per_second_upper_estimate": 0.3681193049604712
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28316.408930467707,
+      "required_pages_per_second_upper_estimate": 0.3632124899689294
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08104677667199667,
-    "producer_p95_seconds": 0.2782366869978432,
-    "gpu_update_seconds": 0.9633802224215462,
-    "producer_batches_per_second": 12.338553623755905,
-    "gpu_batches_per_second": 1.0380117597664675,
-    "producer_headroom": 11.886718534414138,
+    "producer_batch_seconds": 0.24170926782812785,
+    "producer_p95_seconds": 0.27796947300157626,
+    "gpu_update_seconds": 0.9763949965509745,
+    "producer_batches_per_second": 4.137201725798407,
+    "gpu_batches_per_second": 1.0241756702281433,
+    "producer_headroom": 4.039543064791622,
     "sustainable": true,
-    "buffer_seconds": 60.69295401255741,
+    "buffer_seconds": 60.536489786160416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450050.7043022629,
-    "upload_bytes_per_second": 5193181.068173262,
+    "generation_bytes_per_second": 542856.5982570295,
+    "upload_bytes_per_second": 5161246.2219223725,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.57040449906269,
+    "estimated_full_shards_per_hour": 116.48438893111386,
     "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": 1.0684557280001172,
-      "bytes": 6155327,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2350120470000547,
       "bytes": 2680827,
@@ -562,6 +556,12 @@
       "bytes": 6148473,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.581660068997735,
+      "bytes": 6199462,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -311,3 +311,4 @@
 {"step": 751100, "loss": 0.4987703561782837, "lr": 5e-05, "grad_norm": 0.2295593934934252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109939798945, "tokens_per_second": 151038.21410001395, "data_wait_s": 1.4357003237819299e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751200, "loss": 0.5086545944213867, "lr": 5e-05, "grad_norm": 0.22744074776930065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109954326672, "tokens_per_second": 149752.57790120298, "data_wait_s": 1.353399784420617e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751300, "loss": 0.45726361870765686, "lr": 5e-05, "grad_norm": 0.2292745444349719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109969058970, "tokens_per_second": 151617.76236514523, "data_wait_s": 1.2347001757007092e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 751400, "loss": 0.5503573417663574, "lr": 5e-05, "grad_norm": 0.22725561142105227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109983758777, "tokens_per_second": 152447.41646575154, "data_wait_s": 1.6147001588251442e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7736d9dd62bfd7c75332e7dfd0cbf25c5a4f9e2ae0a95b61d20791edb56d83e4
-size 667797
+oid sha256:af02ff79f0a27c7507feb91372b63745a9bf4f2ece44bc3054e9b15a31d7fc97
+size 669915